You are on page 1of 114




Juno starts
watch o
over Jupiter

8 J U LY 2 0 1 6 V O LU M E 3 5 3 I S S U E 6 2 9 5




How can we holistically mine big data?

By N. Prulj and N. Malod-Dognin


Downloaded from on July 8, 2016

Photovoltaics can help to power

remote sensors and controllers
at the edge of the Internet
By R. Haight et al.


We need technology and an ethical
framework for genome-scale
engineering By J. D. Boeke et al.
Not far from downtown Rio de Janeiro, ecologists plot a forest comeback.



Reforms by nations lead biomedical

funder result in shoddy reviews, critics
say By W. Kondro

104 News at a glance




A team of ecologists is recreating a
living rainforest in the heart of the
Olympic city By H. Escobar

National Academies report urges

creation of new national commission on
ethical studies By D. Malakof


FAIR moves ahead despite remaining

budget shortfalls By E. Cartlidge



By E. Ruppel Shell




Interdisciplinary research suggests that

early four-limbed vertebrates relied on
their tails By J. A. Nyakatura







Light-controlled cardiac muscle cells

guide swimming of raybot By E. Pennisi

Metal-organic framework materials

enable efcient separation of similar
gases By J. Y. S. Lin


REPORTS PP. 137 & 141



Meet Marcia McNutt, scientist,

administrator, editor, and now National
Academy of Sciences president

Curtailed ight highlights remaining

challenges of superpressure balloon
technology By P. Monahan

Commercial development could struggle

in a stabilizing market for the worlds
best coolant By E. Hand

By K. McAulife, reviewed by S. Adamo

By A. Robinson,
reviewed by S. DAmico












8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS



The transmembrane domain

of the HIV-1 Env protein

Building human



120 & 154

By M. McNutt

Climate-driven regime shift of a

temperate marine ecosystem
T. Wernberg et al.




Structural basis for membrane

anchoring of HIV-1 envelope spike
J. Dev et al.

By C. Liao et al.




By C. E. Ormsby et al.

Cilia-based ow network in the brain

ventricles R. Faubel et al.




Reengineering chimeric antigen

receptor T cells for targeted therapy of
autoimmune disease C. T. Ellebrecht et al.




133 From Science and other journals

The communities of Science

By Jeremy Berg




Thermally stable single-atom platinumon-ceria catalysts via atom trapping

J. Jones et al.

Asymmetric division of clonal

muscle stem cells coordinates
muscle regeneration in vivo
D. B. Gurevich et al.


137 A metal-organic frameworkbased
splitter for separating propylene
from propane A. Cadiau et al.

141 Pore chemistry and size control

in hybrid porous materials for
acetylene capture from ethylene
X. Cui et al.


Tail use improves performance on soft
substrates in models of early vertebrate
land locomotors B. McInroe et al.

Phototactic guidance of a tissueengineered soft-robotic ray
S.-J. Park et al.


Higher-order organization of complex
networks A. R. Benson et al.





Copper-catalyzed asymmetric addition

of olen-derived nucleophiles to ketones
Y. Yang et al.

S-Acylation of the cellulose synthase

complex is essential for its plasma
membrane localization M. Kumar et al.


The questions that opened doors
By Carlos A. Aguilar-Trigueros

A robotic stingray made
up of rat heart cells,
gold, and a polymer
commonly used in
breast implants. The
robot ray mimics the
form and function
of a swimming
stingray, but with a
twist: Genetically altered rat cells allow
the creature to sense and swim toward
light. See pages 110 and 158. For more
on the process behind the cover image,
Photo: Ken Richardson
Science Staf .............................................. 102
New Products ............................................. 185
Science Careers ......................................... 186

SCIENCE (ISSN 0036-8075) is published weekly on Friday, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Periodicals mail
postage (publication No. 484460) paid at Washington, DC, and additional mailing ofces. Copyright 2016 by the American Association for the Advancement of Science. The title SCIENCE is a registered trademark of the AAAS. Domestic
individual membership and subscription (51 issues): $165 ($74 allocated to subscription). Domestic institutional subscription (51 issues): $1522; Foreign postage extra: Mexico, Caribbean (surface mail) $55; other countries (air assist
delivery) $89. First class, airmail, student, and emeritus rates on request. Canadian rates with GST available upon request, GST #1254 88122. Publications Mail Agreement Number 1069624. Printed in the U.S.A.
Change of address: Allow 4 weeks, giving old and new addresses and 8-digit account number. Postmaster: Send change of address to AAAS, P.O. Box 96178, Washington, DC 200906178. Single-copy sales: $15.00 current issue, $20.00
back issue prepaid includes surface postage; bulk rates on request. Authorization to photocopy material for internal or personal use under circumstances not falling within the fair use provisions of the Copyright Act is granted by AAAS
to libraries and other users registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided that $35.00 per article is paid directly to CCC, 222 Rosewood Drive, Danvers, MA 01923. The identifcation code
for Science is 0036-8075. Science is indexed in the Readers Guide to Periodical Literature and in several specialized indexes.


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016




1200 New York Avenue, NW, Washington, DC 20005

Clarendon House, Clarendon Road, Cambridge, UK CB2 8FH

Gary King, Harvard University, Susan M. Rosenberg, Baylor College of Medicine,

Ali Shilatifard, Northwestern University Feinberg School of Medicine

BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)

Editor-in-Chief Jeremy Berg

Executive Editor Monica M. Bradford
News Editor Tim Appenzeller
Managing Editor, Research Journals Katrina L. Kelner
Deputy Editors Lisa D. Chong, Andrew M. Sugden(UK), Valda J. Vinson, Jake S. Yeston

Research and Insights

DEPUTY EDITOR, EMERITUS Barbara R. Jasny SR. EDITORS Caroline Ash(UK), Gilbert J. Chin, Julia Fahrenkamp-Uppenbrink(UK),

Pamela J. Hines, Stella M. Hurtley(UK), Paula A. Kiberstis, Marc S. Lavine(Canada), Kristen L. Mueller, Ian S. Osborne(UK),
Beverly A. Purnell, L. Bryan Ray, Guy Riddihough, H. Jesse Smith, Jelena Stajic, Peter Stern(UK), Phillip D. Szuromi, Sacha
Vignieri, Brad Wible, Nicholas S. Wigginton, Laura M. Zahn ASSOCIATE EDITORS Brent Grocholski, Keith T. Smith ASSOCIATE BOOK
PRODUCTION EDITORS Jeffrey E. Cook, Chris Filiatreau, Cynthia Howe, Barbara P. Ordway, Catherine Wolner SR. EDITORIAL COORDINATORS
Carolyn Kyle, Beverly Shields EDITORIAL COORDINATORS Aneera Dobbins, Joi S. Granger, Lisa Johnson, Maryrose Madrid, Anita Wynn
PUBLICATIONS ASSISTANTS Jeffrey Hearn, Dona Mathieu, Le-Toya Mayne Flood, Shannon McMahon, Scott Miller, Jerry Richardson, Alice
Whaley(UK), Brian White EXECUTIVE ASSISTANT Anna Bashkirova ADMINISTRATIVE SUPPORT Janet Clements(UK), Lizanne Newton(UK)

David Grimm, David Malakoff, Leslie Roberts CONTRIBUING EDITOR Martin Enserink(Europe) SR. CORRESPONDENTS Daniel Clery(UK),
Jeffrey Mervis, Elizabeth Pennisi NEWS WRITERS Adrian Cho, Jon Cohen, Jennifer Couzin-Frankel, Carolyn Gramling, Eric Hand,
Jocelyn Kaiser, Catherine Matacic, Kelly Servick, Robert F. Service, Erik Stokstad(Cambridge, UK), Emily Underwood INTERNS
Patrick Monahan CONTRIBUTING CORRESPONDENTS John Bohannon, Warren Cornwall, Ann Gibbons, Mara Hvistendahl, Sam Kean,

Eli Kintisch, Kai Kupferschmidt(Berlin), Andrew Lawler, Christina Larson(Beijing), Mitch Leslie, Charles C. Mann, Eliot Marshall,
Virginia Morell, Dennis Normile(Shanghai), Heather Pringle, Tania Rabesandratana(London), Gretchen Vogel(Berlin), Lizzie
Wade(Mexico City) CAREERS Donisha Adams, Rachel Bernstein(Editor) COPY EDITORS Julia Cole, Dorie Chevlen, Jennifer Levin

Executive Publisher Rush D. Holt

Publisher Bill Moran
Chief Digital Media Officer Rob Covey
Latasha Russell MANAGER, DATA ENTRY Mickie Napoleoni DATA ENTRY SPECIALISTS Brenden Aquilino, Fiona Giblin MARKETING
ANALYST Judy Lillibridge
DEVELOPERS Ryan Jensen, Jimmy Marks, Brandon Morrison
DIGITAL/PRINT Rebecca Doshi SENIOR CONTENT SPECIALISTS Steve Forrester, Antoinette Hodal, Lori Murphy, Anthony Rosen
CONTENT SPECIALISTS Jacob Hedrick, Kimberley Oster
DESIGN DIRECTOR Beth Rakouskas DESIGN EDITOR Marcy Atarod SENIOR DESIGNERS Garvin Grulln, Chrystal Smith
Adarsh Sandhu + 81532-81-5142 EAST COAST/E. CANADA Laurie Faraday: 508-747-9395, FAX 617-507-8189 WEST COAST/W.
CANADA Lynne Stickrod: 415-931-9782, FAX 415-520-6940 MIDWEST Jeffrey Dembksi: 847-498-4520 x3005, Steven Loerch: 847-498-4520 x3006
UK EUROPE/ASIA Roger Goncalves: TEL/FAX +41 43 243 1358 JAPAN Katsuyoshi Fukamizu(Tokyo): +81-3-3219-5777 CHINA/
TAIWAN Ruolei Wu: +86-186 0082 9345,
WORLDWIDE ASSOCIATE DIRECTOR OF SCIENCE CAREERS Tracy Holmes: +44 (0) 1223 326525, FAX +44 (0) 1223 326532 tholmes@science-int.
CLASSIFIED U.S. SALES Tina Burks: 202-326-6577, Nancy Toema: 202-326-6578 EUROPE/ROW SALES Sarah Lelarge SALES
ASSISTANT Kelly Grace JAPAN Hiroyuki Mashiki(Kyoto): +81-75-823-1109 CHINA/TAIWAN Ruolei Wu: +86-186 0082 9345 rwu@ MARKETING MANAGER Allison Pritchard MARKETING ASSOCIATE Aimee Aponte
David Evans Shaw CHIEF EXECUTIVE OFFICER Rush D. Holt BOARD Cynthia M. Beall, May R. Berenbaum, Carlos J. Bustamante,

Stephen P.A. Fodor, Claire M. Fraser, Michael S. Gazzaniga, Laura H. Greene, Elizabeth Loftus, Mercedes Pascual
SUBSCRIPTION SERVICES For change of address, missing issues, new orders and renewals, and payment questions: 866-434-AAAS (2227) or 202-326-6417, FAX 202-8421065. Mailing addresses: AAAS, P.O. Box 96178, Washington, DC 20090-6178 or AAAS Member Services, 1200 New York Avenue, NW, Washington, DC 20005
INSTITUTIONAL SITE LICENSES 202-326-6730 REPRINTS: Author Inquiries 800-635-7181 COMMERCIAL INQUIRIES 803-359-4578 PERMISSIONS 202-326-6765, AAAS Member Services 202-326-6417 or
Science serves as a forum for discussion of important issues related to the advancement of science by publishing material on which a consensus has been reached as
well as including the presentation of minority of conflicting points of view. Accordingly, all articles published in Scienceincluding editorials, news and comment, and book
reviewsare signed and reflect the individual views of the authors and not official points of view adopted by AAAS or the institutions with which the authors are affiliated.
INFORMATION FOR AUTHORS See pages 624 and 625 of the 5 February 2016 issue or access


Adriano Aguzzi, U. Hospital Zrich

Takuzo Aida, U. of Tokyo
Leslie Aiello, Wenner-Gren Foundation
Judith Allen, U. of Edinburgh
Sonia Altizer, U. of Georgia
Sebastian Amigorena, Institut Curie
Kathryn Anderson, Memorial
Sloan-Kettering Cancer Center
Meinrat O. Andreae, Max-Planck Inst. Mainz
Paola Arlotta, Harvard U.
Johan Auwerx, EPFL
David Awschalom, U. of Chicago
Clare Baker, University of Cambridge
Nenad Ban, ETH Zurich
Jordi Bascompte, University of Zurich
Franz Bauer, Instituto de Astrofisica
Ray H. Baughman, U. of Texas, Dallas
David Baum, U. of Wisconsin
Carlo Beenakker, Leiden U.
Kamran Behnia, ESPCI-ParisTech
Yasmine Belkaid, NIAID, NIH
Philip Benfey, Duke U.
May Berenbaum, U. of Illinois
Gabriele Bergers, U. of California, San Francisco
Bradley Bernstein, Massachusettes General Hospital
Peer Bork, EMBL
Bernard Bourdon, Ecole Normale Suprieure de Lyon
Chris Bowler, cole Normale Suprieure
Ian Boyd, U. of St. Andrews
Emily Brodsky, U. of California, Santa Cruz
Ron Brookmeyer, U. of California Los Angeles (S)
Christian Bchel, U. Hamburg-Eppendorf
Joseph A. Burns, Cornell U.
Carter Tribley Butts, U. of California, Irvine
Gyorgy Buzsaki, New York U. School of Medicine
Blanche Capel, Duke U.
Mats Carlsson, U. of Oslo
Ib Chorkendorff, U. of Denmark
David Clapham, Childrens Hospital Boston
Joel Cohen, Rockefeller U., Columbia U.
James J. Collins, MIT
Robert Cook-Deegan, Duke U.
Lisa Coussens, Oregon Health & Science U.
Alan Cowman, Walter & Eliza Hall Inst.
Robert H. Crabtree, Yale U.
Roberta Croce, Vrije Universiteit
Janet Currie, Princeton U.
Jeff L. Dangl, U. of North Carolina
Tom Daniel, U. of Washington
Frans de Waal, Emory U.
Stanislas Dehaene, Collge de France
Robert Desimone, MIT
Claude Desplan, New York U.
Dennis Discher, U. of Pennsylvania
Gerald W. Dorn II, Washington U. School of Medicine
Jennifer A. Doudna, U. of California, Berkeley
Bruce Dunn, U. of California, Los Angeles
William Dunphy, Caltech
Christopher Dye, WHO
Todd Ehlers, U. of Tuebingen
David Ehrhardt, Carnegie Inst. of Washington
Tim Elston, U. of North Carolina at Chapel Hill
Gerhard Ertl, Fritz-Haber-Institut, Berlin
Barry Everitt, U. of Cambridge
Ernst Fehr, U. of Zurich
Anne C. Ferguson-Smith, U. of Cambridge
Michael Feuer, The George Washington U.
Toren Finkel, NHLBI, NIH
Kate Fitzgerald, U. of Massachusetts
Peter Fratzl, Max-Planck Inst.
Elaine Fuchs, Rockefeller U.
Daniel Geschwind, UCLA
Karl-Heinz Glassmeier, TU Braunschweig
Ramon Gonzalez, Rice U.
Julia R. Greer, Caltech
Elizabeth Grove, U. of Chicago
Nicolas Gruber, ETH Zurich
Kip Guy, St. Judes Childrens Research Hospital
Taekjip Ha, U. of Illinois at Urbana-Champaign
Wolf-Dietrich Hardt, ETH Zurich
Christian Haass, Ludwig Maximilians U.
Sharon Hammes-Schiffer, U. of Illinois at
Michael Hasselmo, Boston U.
Martin Heimann, Max-Planck Inst. Jena
Yka Helariutta, U. of Cambridge
James A. Hendler, Rensselaer Polytechnic Inst.
Janet G. Hering, Swiss Fed. Inst. of Aquatic
Science & Technology
Kai-Uwe Hinrichs, U. of Bremen
David Hodell, U. of Cambridge
Lora Hooper, UT Southwestern Medical Ctr. at Dallas
Tamas Horvath, Yale University
Raymond Huey, U. of Washington
Fred Hughson, Princeton U.
Auke Ijspeert, EPFL Lausanne
Stephen Jackson, USGS and U. of Arizona
Steven Jacobsen, U. of California, Los Angeles
Kai Johnsson, EPFL Lausanne
Peter Jonas, Inst. of Science & Technology (IST) Austria
Matt Kaeberlein, U. of Washington
William Kaelin Jr., Dana-Farber Cancer Inst.
Daniel Kahne, Harvard U.
Daniel Kammen, U. of California, Berkeley
Abby Kavner, U. of California, Los Angeles
Masashi Kawasaki, U. of Tokyo

V. Narry Kim, Seoul National U.

Joel Kingsolver, U. of North Carolina at Chapel Hill
Robert Kingston, Harvard Medical School
Etienne Koechlin, Ecole Normale Suprieure
Alexander Kolodkin, Johns Hopkins U.
Thomas Langer, U. of Cologne
Mitchell A. Lazar, U. of Pennsylvania
David Lazer, Harvard U.
Thomas Lecuit, IBDM
Virginia Lee, U. of Pennsylvania
Stanley Lemon, U. of North Carolina at Chapel Hill
Ottoline Leyser, Cambridge U.
Wendell Lim, U.C. San Francisco
Marcia C. Linn, U. of California, Berkeley
Jianguo Liu, Michigan State U.
Luis Liz-Marzan, CIC biomaGUNE
Jonathan Losos, Harvard U.
Ke Lu, Chinese Acad. of Sciences
Christian Lscher, U. of Geneva
Laura Machesky, CRUK Beatson Inst.
for Cancer Research
Anne Magurran, U. of St. Andrews
Oscar Marin, CSIC & U. Miguel Hernndez
Charles Marshall, U. of California, Berkeley
C. Robertson McClung, Dartmouth College
Graham Medley, U. of Warwick
Tom Misteli, NCI
Yasushi Miyashita, U. of Tokyo
Mary Ann Moran, U. of Georgia
Richard Morris, U. of Edinburgh
Alison Motsinger-Reif, NC State U. (S)
Thomas Murray, The Hastings Center
Daniel Neumark, U. of California, Berkeley
Kitty Nijmeijer, U. of Twente
Helga Nowotny, European Research Advisory Board
Ben Olken, MIT
Joe Orenstein, U. of California
Berkeley & Lawrence Berkeley National Lab
Harry Orr, U. of Minnesota
Pilar Ossorio, U. of Wisconsin
Andrew Oswald, U. of Warwick
Margaret Palmer, U. of Maryland
Steve Palumbi, Stanford U.
Jane Parker, Max-Planck Inst.
of Plant Breeding Research
Giovanni Parmigiani, Dana-Farber Cancer Inst. (S)
John H. J. Petrini, Memorial Sloan-Kettering
Cancer Center
Samuel Pfaff, Salk Institute for Biological Studies
Joshua Plotkin, U. of Pennsylvania
Albert Polman, FOM Institute AMOLF
Philippe Poulin, CNRS
Jonathan Pritchard, Stanford U.
David Randall, Colorado State U.
Felix Rey, Institut Pasteur
Trevor Robbins, U. of Cambridge
Jim Roberts, Fred Hutchinson Cancer Research Ctr.
Barbara A. Romanowicz, U. of California,Berkeley
Amy Rosenzweig, Northwestern University
Mike Ryan, U. of Texas, Austin
Mitinori Saitou, Kyoto U.
Shimon Sakaguchi, Kyoto U.
Miquel Salmeron, Lawrence Berkeley National Lab
Jrgen Sandkhler, Medical U. of Vienna
Alexander Schier, Harvard U.
Vladimir Shalaev, Purdue U.
Robert Siliciano, Johns Hopkins School of Medicine
Denis Simon, Arizona State U.
Uri Simonsohn, U. of Pennsylvania
Alison Smith, John Innes Centre
Richard Smith, U. of North Carolina (S)
John Speakman, U. of Aberdeen
Allan C. Spradling, Carnegie Institution
of Washington
Jonathan Sprent, Garvan Inst. of Medical Research
Eric Steig, U. of Washington
Paula Stephan, Georgia State U. and National Bureau
of Economic Research
Molly Stevens, Imperial College London
V. S. Subrahmanian, U. of Maryland
Ira Tabas, Columbia U.
Sarah Teichmann, Cambridge U.
John Thomas, North Carolina State U.
Shubha Tole, Tata Institute of Fundamental Research
Christopher Tyler-Smith, The Wellcome Trust
Sanger Inst.
Herbert Virgin, Washington U.
Bert Vogelstein, Johns Hopkins U.
Cynthia Volkert, U. of Gttingen
David Wallach, Weizmann Inst. of Science
Ian Walmsley, U. of Oxford
Jane-Ling Wang, U. of California, Davis (S)
David A. Wardle, Swedish U. of Agric. Sciences
David Waxman, Fudan U.
Jonathan Weissman, U. of California, San Francisco
Chris Wikle, U. of Missouri (S)
Ian A. Wilson, The Scripps Res. Inst. (S)
Timothy D. Wilson, U. of Virginia
Rosemary Wyse, Johns Hopkins U.
Jan Zaanen, Leiden U.
Kenneth Zaret, U. of Pennsylvania School of Medicine
Jonathan Zehr, U. of California, Santa Cruz
Len Zon, Childrens Hospital Boston
Maria Zuber, MIT


David Bloom, Harvard U., Samuel Bowring, MIT, Angela Creager, Princeton U.,
Richard Shweder, U. of Chicago, Ed Wasserman, DuPont SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


y father was a mathematician. When I was a

teenager, I talked to him about careers. I love
math, even fairly esoteric stuff. He surprised
me by saying, Math is something to pursue
only if you cannot imagine doing anything else.
You can follow other careers and still do math,
but pure math can be very isolating; only a
few people in the world are likely to really understand
what you are working on.
I was surprised that he was
discouraging me from his
career path. But wait, what?
He was also implying that
science was a social activity
in an essential way.
I took his advice and pursued chemistry, driven in
part by my interest in geometry and three-dimensional
structures. Over my career,
I have moved from chemistry to structural biology
and crystallography, to synthetic inorganic chemistry,
back to structural biology, to
biophysics, to biochemistry,
to informatics, to computational biology, and to personalized medicine. Along the
way, I also have had great opportunities to lead scientific
organizations. I helped build
a department of biophysics
and directed the most fundamental science-focused institute at the U.S. National Institutes of Health (NIH), with programs spanning the basic
biomedical and related sciences, key clinical areas such as
anesthesiology, and research training. With regard to the
clinical areas, my expertise has been greatly bolstered by
home schooling in clinical research by my wife, Wendie
Berg, an accomplished breast imaging researcher and clinician who has run several major clinical trials.
While at NIH, I had many intense discussions with a
colleague about the social science research relevant to
developing inclusive training programs. He would frequently challenge me with results that did not fit my
intuition. He had the habit of adding aphorisms to his
email signature block, and one of my favorites was, Dont
believe everything you think. This directive nicely captures the power of the scientific method. If you have an
idea, you can collect data and analyze them to determine

whether your observations are consistent with your hypothesis or refute it. I, like most practicing scientists,
have had my eyes opened when an attractive hypothesis
was demolished by an incisive experiment.
Returning to my fathers comment, science is, indeed,
a profoundly social activity. Many hypotheses come not
from single individuals but from groups with individuals
at different career stages and with diverse interests, discussing previous results and
ideas. Small or large teams
of researchers perform many
experiments. Other scientists then review the results,
critically examining the experiments, analyses, and
interpretation. All of this is
conducted in a highly competitive environment where
personal satisfaction, funding, recognition, and career
stability often depend on being the first to make an advance, creating tension with
collaborative efforts.
It is daunting to be added
to the list of incredibly impressive individuals who
have previously served as editor-in-chief. Most recently,
Marcia McNutt has led the
Science journals admirably
over her relatively short tenure. She is leaving only because her accomplishments
led to another amazing opportunity, and the scientific
community is fortunate to have her as president of the
U.S. National Academy of Sciences.
The position of the editor-in-chief of Science and the
other journals of the Science family presents both a tremendous opportunity and an intimidating responsibility.
Science is a key tool in addressing the pressing problems
facing humankind and our planet. To make progress, scientists must engage with the public, policy-makers, and
many other groups. Scientists must communicate their
results fairly and listen to the perspectives of others. Science serves as a premier journal for presenting the most
exciting scientific results from a wide range of disciplines
and is a key forum for relevant news and discussions of
policy. I look forward to working with the communities
of Science.

Jeremy Berg
Science Journals
Email: jberg@


science is, indeed, a profoundly

social activity.

Jeremy Berg



15 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

The communities of Science


Under the law, it says you must report. If you dont

report, the law says you shouldnt get funding.

Vice President Joe Biden, at a 29 June summit officially launching the White Houses
moonshot to fast-track cancer research, promising to crack down on research
institutions that dont report their clinical trial data.

Jupiter has a new orbiter

NEONs price tag goes up
| The National
Science Foundation (NSF) has agreed to
spend $35.5 million more to complete its
National Ecological Observatory Network
(NEON), raising its total price to $469
million. The money is expected to allow
the troubled project, already reduced in
scope and delayed by a year, to begin full
operations at all 81 sites by the end of
2017. The additional construction money
comes from a pot earmarked for maintenance and operations for NEON; that
money had not yet been spent because the
project is behind schedule. But the project
could still siphon away money from NSFs
existing biology research program, because
NEONs new contractor, Battelle Memorial
Institute, hasnt yet worked up a detailed
proposal for how much the continent-wide
observatory will cost to operate. NSFs
WA S H I N GT O N , D.C .


current estimate of $60 million to $80

million is not reliable, says Maria Zuber,
chair of the National Science Board in
Arlington, Virginia, which oversees NSF
and approved the additional spending last
week. If operating costs become an issue,

then we may be talking about another

descopingaltering the projects objectives, Zuber says. But lets get it built,
and then we can look closely at managing
the impact of NEON on the rest of the bio

A research tower at NEONs Front Royal, Virginia, site.

Call for research integrity office

| Like many countries, France
started giving attention to the issue of scientific misconduct in the 1990s, culminating
in a 2015 charter for scientific integrity
signed by several major national research
organizations and the countrys Conference
of University Presidents. But compared with
other European and North American countries, the countrys research organizations
and universities have been lagging behind
in putting in place appropriate measures,
suggests a 29 June report commissioned
by Thierry Mandon, Frances state secretary for higher education and research.

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016



Juno (here in an artists rendition) signaled its

safe arrival at Jupiter with three simple tones.

fter a 5-year journey, NASAs latest planetary

probe, Juno, safely completed a tricky maneuver on 4 July to enter into orbit around the
solar systems biggest planet. By the end of
August, the craft will settle into a 14-day elliptical orbit that will bring it within 4000 kilometers of Jupiters outer veil of clouds, closer than any
spacecraft has gone before. Juno will peer through
the clouds to study the planets structure; scientists
hope it will answer questions such as how far down
the surface stripes and swirling storms go, what is
the source of Jupiters intense magnetic field, and
whether the planet has a rocky core. Theorists believe
that, as the solar systems largest resident, Jupiter
holds the key to understanding how the whole body
of planets formed around the protostar that became
our sun. The probe will circle the planet pole to pole
33 times over the next year and a half before plunging into the clouds at the end of its mission to prevent
an accidental collision with one of Jupiters moons,
which could contaminate it with earthly microbes.

Mexicos wind farms, like

this one on the Isthmus
of Tehuantepec, face
community opposition.

Charitys open access push

| The Wellcome Trust, the
U.K.-based charity that has become one of
the biggest nongovernmental funders of
biomedical research, will launch its own
open-access online journal this autumn.
Publication will be limited to the thousands
of scientists working on research funded
by a Wellcome grant, and it will be free
for authors as well as for readersthe $24
billion charity is covering the costs of the
journals software and online platform. Peer
review will occur postpublication, the charity said in a statement. The move has excited
many critics of traditional scientific publishing. This really is a potential game changer
for a major funder to be taking control of
the research output, says Paul Ginsparg, the
Cornell University physicist who founded
arXiv, the massive online scientific preprint
server. Wellcome previously launched the
open access journal eLife in 2012, in collaboration with the Howard Hughes Medical
Institute and the Max Planck Society.

Dawn, New Horizons updates

| NASA announced
1 July that it is denying a third act to its
Dawn spacecraft, which has so far explored
the two largest bodies in the asteroid belt,
Vesta and Ceres. Rather than visiting a
150-kilometer-wide asteroid known as
145 Adeona in 2019, as proposed by its
mission leaders, the spacecraft will remain
in orbit around Ceres, the administration
announced, following the recommendations of a senior review panel. Chris Russell,
the mission principal investigator at the
University of California, Los Angeles, says he
was surprised and disappointed to hear of
the decision. We thought that everyone we
had talked to about this plan was enthusiastic about it, he says. I had no negative vibes
until this particular moment. Meanwhile,


WA S H I N GT O N , D.C .

Mexicos clean energy challenge

ast week, Mexico, Canada, and the United States signed an agreement to commit
to generating 50% of their electricity from clean energy sources by 2025. That
may present a particular challenge for Mexico, which relies on fossil fuels for 78%
of its electricity (compared with 25% in Canada and 67% in the United States).
I think its doable. Its good to have such an ambitious goal, says Juan BezauryCreel of the Nature Conservancy, who is based in Mexico City. A controversial energy
reform enacted in Mexico between 2013 and 2015 opened the market to decentralized
renewable energy providers and will require businesses to pay if they exceed their
allotment of energy from fossil fuels. Still, many energy policy experts wonder whether
enforcement will be up to snuff in a country where corruption and impunity thrive.
Mexico has also committed to tripling its wind power capacity by 2018. The Isthmus
of Tehuantepec in the state of Oaxaca in Mexicothe thin strip of land between the
Atlantic and Pacific Oceansis one of the windiest places in the world and already
hosts about 1600 turbines. But several ambitious wind projects have been delayed
or canceled by conflicts with indigenous communities over land rights.

NASA said it will support a mission extension for New Horizons, the spacecraft that
flew past Pluto in July 2015. The spacecraft
will now pursue 2014 MU69, an icy Kuiper
belt object about 30 kilometers across,
for a rendezvous on 1 January 2019.

golden rice, a genetically engineered

variety of rice intended to protect against
vitamin A deficiency in children; it is
enriched with b-carotene, a precursor of
vitamin A. Greenpeace posted a response
to the letter on its website, denying that its
opposition is responsible for the failure of
golden rice to come to market.

Nobelists: Drop anti-GMO stance

| More than 100 Nobel
laureates have signed a letter, publicized
last week in The Washington Post, that
strongly chastises the Washington, D.C.
based environmental group Greenpeace
for its opposition to genetically modified
organisms, stating that the group has
misrepresented their risks, benefits, and
impacts and suggesting that such opposition is based on emotion and dogma
rather than scientific data. In particular,
the letter calls on Greenpeace to cease
and desist its campaign against so-called
WA S H I N GT O N , D.C .


On 27 June, Science won four 2016
EXCEL Awards, which recognize
excellence in art and design for nonprofit
media. The awarded work included
cover designs for the 6 November and
4 December 2015 issues and the design
for the 5 June 2015 Feature on isolated
tribes. Science was also recognized for
design excellence for three consecutive
issues: 4 December, 11 December, and
18 December 2015.

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

Research organizations in the survey, and in

particular the French National Institute of
Health and Medical Research, generally had
stronger procedures in place than universities, and most institutions had a policy for
paper authorship, the report found. But
there were partial, and sometimes even
nonexistent procedures reporting and
investigating alleged misconduct at many
institutions. The report calls for a national
strategy and the creation of a French Office
for Research Integrity to monitor questionable research practices and support
institutions in their handling of misconduct.
It also suggests handing out research grants
only to institutions that have a scientific
integrity policy.

Downloaded from on July 8, 2016



Panel slams plan for human research rules

National Academies report urges creation of new national commission on ethical studies

n a surprise development certain to fuel

a long-running controversy, a prominent
science advisory panel is calling on the
U.S. government to abandon a nearly
finished update to rules on protecting
human research participants. It should
wait for a new high-level commission, created by Congress and the president, to recommend improvements and then start over,
the panel says.
Policy insiders say the recommendation,
made 29 June by a committee of the National Academies of Sciences, Engineering,
and Medicine that is examining ways to
reduce the regulatory burden on academic
scientists, is the political equivalent of a
comic book hero trying to step in front of a
speeding train in a bid to prevent a wreck.
Its not clear, however, whether the panel
will succeed in stopping the regulatory expressor just get run over. Both the Obama
administration, which has been pushing to
complete the new rules this year, and lawmakers in Congress would need to back the
haltand so far theyve been silent.
Still, many researchers and university
groups are thrilled with the panels recommendation, noting that they have repeatedly objected to some of the proposed rule
changes as unworkable, but with little apparent impact. Weve been saying for some
time that there is no rush to issue this rule

by the end of this administration, says Lisa

ally support the intent of the changes,
Nichols, director of research and regulatory
they have blasted many of the proposals as
reform at the Council on Governmental
overly burdensome and counterproductive.
Relations in Washington, D.C., which repOne lightning rod has been a proposal to
resents universities and other research inallow researchers to use anonymized tissue
stitutions. The panel offers more support
samples only if they come from people who
for the idea that we need further discussion
have given explicit written consenta plan
among experts and stakeholders before we
critics say would create a paperwork nightmove ahead.
mare and place many useful samples off
At the center of the debate is the so-called
limits (Science, 20 May, p. 878). ResearchCommon Rule, which sets
ers also have argued that
ethical standards for fedit isnt clear how key
erally funded researchers
components might work,
who work with human
including what kinds of
subjects. The rule spells
social science research
out consent rules and
such as analyses of
governs how researchtweetsmight be exempt
ers can use data and
from consent rules.
Committee on Federal Research
tissue specimens from
The National AcadRegulations and Reporting
humans, alive or dead.
emies panel, which was
Officials formulated the
created to examine a
rule in 1981, with help from an influential
range of federal regulations that affect
1978 study known as the Belmont Report,
academics, echoes those concerns. The
which was produced by a presidential comCommon Rule proposal is marred by omismission. Since then, the rule has undergone
sions, the absence of essential elements,
periodic tweaks, but in 2009 more than a
and a lack of clarity, the report states, and
dozen federal agencies launched a major efshould be withdrawn. The inadequacies,
fort to modernize it, concluding that shiftit says, signal a pressing need for a coming ethical concerns and new biomedical
prehensive review of the nations ethical,
and digital technologies had rendered it
legal, regulatory and institutional frameoutdated. Late last year, the agencies asked
works for protecting research subjects. To
for public comment on a draft rule that proconduct that review, the panel calls for a
posed scores of changes.
national expert panel modeled on the one
Although academic researchers generthat produced the Belmont Report.

The Common Rule

proposal is marred
by omissions SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


By David Malakoff



The idea for the new commissionand

the Common Rule haltlargely arose from
the committees realization that the past
consensus over how we should regulate human research is broken, says panelist Arturo
Casadevall, a microbiologist at Johns Hopkins University in Baltimore, Maryland. Now,
said Larry Faulkner, chair of the committee
and president emeritus at the University of
Texas, Austin, in a statement, Congress and
the administration have an opportunity for a
course correction.
Neither administration officials nor lawmakers, however, were immediately willing
to buy in. A spokesperson for the Department of Health and Human Services, which
is helping lead the reform effort, wrote in
an email that officials planned to read the
report, but suggested that the reform effort
may continue. If a final rule is issued this
fall, it will be the first time in 25 years that
the regulations governing research using
human participants has been updated, the
spokesperson said.
Senator Lamar Alexander (RTN), who
requested the panels report and chairs the
Senates health committee, is still reviewing
the recommendation, his press secretary
Louie Brogdon told Science. But he says that
Alexander believes the Common Rule must
strike a balance between allowing researchers access to biospecimens, such as blood or
tissue, and ensuring the privacy of patients.
Alexanders committee could be a potential
launching pad for legislation creating the
proposed commission.
Many critics of the Common Rule draft
had essentially given up hope that the administration would back away from its push
to issue a final rule this year. But the panels
unexpected move has raised a glimmer of
hope, says Emma Meagher, an associate vice
provost for human research at the University
of Pennsylvania. The recommendation cant
be ignored given the profile of this group and
[its] thoughtfulness, predicts Elisa Hurley,
executive director of Public Responsibility in
Medicine and Research in Boston.
That doesnt mean the administration will
agree with the proposal, Hurley adds. One
possibility, observers say, is that the administration will withdraw particularly controversial elements and push ahead with the
rest. Such a move could still leave room
for a new commission to really think logically about research regulations in general,
Meagher says. And Casadevall says, we are
really in need of a comprehensive, fresh look
at the problem. j
With reporting by Dianne Lugo.


Long-delayed nuclear center

looks set for construction
FAIR moves ahead despite remaining budget shortfalls
By Edwin Cartlidge

hat was supposed to be a global

beehive of nuclear physics is, for
now, still a large, muddy field
outside the German city of Darmstadt. The Facility for Antiproton
and Ion Research (FAIR), an extension of the GSI Helmholtz Centre for
Heavy Ion Research planned by a collaboration of eight European Union countries
plus Russia and India, was first expected
to open in 2009, and then in 2015. But a
series of planning errors and other problems pushed the price tag up to some 1.7
billionan overall rise of about 75%and
stalled construction.

commit their share of the missing cash, including Russia, which had agreed to bear
about 18% of FAIRs total construction cost,
the second largest contribution after Germanys 70%. A statement issued by the FAIR
council is short on details about the exact
financial situation.
At FAIR, an 1100-meter-circumference
synchrotron housed in a circular tunnel will
propel extremely intense ion beams to very
high energies and direct them to storage
rings and detectors located above ground,
allowing scientists to study everything from
the origin of chemical elements to the cosmic imbalance between matter and antimatter (Science, 2 November 2007, p. 738).
In the first proposal, in 2001, FAIR was

The site reserved for the Facility for Antiproton and Ion Research near Darmstadt, Germany, pictured in 2014.

Now, however, FAIR may finally get built.

At a council meeting on 27 and 28 June, the
partner countries concluded that they have
enough money to cover a 320 million budget gap that came to light in 2014, and they
will now seek building permits from the
German government. If all goes well, FAIR
could be completed in 2025. It is absolutely
great news, says Carsten Welsch, a physicist at the University of Liverpool in the
United Kingdom, which became an associate partner of FAIR in 2013. The scientific
community has been waiting for years for
this project to go ahead.
Yet not all of the financial problems have
been solved. Some countries still have to


predicted to cost 675 million and to start

operations in 2009. That was too optimistic.
First, design additions, an increase in the
cost of raw materials, and inflation upped the
price to some 1.2 billion. Then, an additional
roughly 100 million was needed to stabilize the unexpectedly sandy ground at the
location. In 2009, the nine full-partner
countries decided to commit just over
1 billion in 2005 prices for a scaled-back
design with fewer instruments, to open
by 2015.
Then the new 320 million hole emerged,
mostly due to the introduction of tighter
fire regulations in Germany that had major
design consequences, says Ulrich Wiedner,
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

Debate surrounds rules to protect human research

subjects, such as this person in a sleep study.


Edwin Cartlidge is a writer based in Rome.


Downloaded from on July 8, 2016


Titanic balloon sets record

and tantalizes scientists
Curtailed flight highlights remaining challenges of
superpressure balloon technology
By Patrick Monahan

he latest and largest pressurized balloon to be launched by NASA has set

a record for endurance: the longest
midlatitude flight by a large scientific
balloon. Packing 532,000 cubic meters
of helium and measuring 114 meters
in diameter, the balloon circled the Southern
Hemisphere for 46 days, lofting a gamma ray
telescope to the edges of space. Nightly dips
in altitude forced a premature end to the voyage on 2 July, but the flight still marks a milestone in NASAs efforts to develop so-called
superpressure balloons as a low-cost alternative to satellites.
For decades, conventional zero-pressure
balloons have given researchers a highaltitude platform for studying atmospheric
chemistry, the cosmic microwave background, and many other phenomena. But
at temperate latitudes, the endurance of
conventional balloons is limited. During the
daytime, sunlight heats the helium, causing
the gas to expand and leak. At night, the balloon cools and must drop ballast to avoid
drifting too low. Zero-pressure balloons can
only achieve long flights during summertime
near the poles, when constant daylight allows
them to stay afloat for weeks at a time.
Superpressure balloons promise to bring
that endurance to temperate latitudes, opening new phenomena to observation. Their
helium is pressurized; because their volume

doesnt change as they are heated or cooled

over the course of the day, they can remain
at a constant altitude and dont need to shed
gas or ballast. Over the past decade, NASA
has launched ever larger and more ambitious
superpressure balloons, clocking midlatitude
flights of up to 32 days.
On 17 May, the agency launched its latest
model from Wanaka, New Zealand. But after

Going the distance

NASAs longest heavy-lift flight spanned 55 days in
201213, but the Super-TIGER zero-pressure balloon
traced a tight path (in blue) around the South Pole. The
latest balloon set a midlatitude duration record.
Superpressure balloon

Caman, PE


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


a nuclear physicist at Ruhr University Bochum in Germany and former spokesperson

of Antiproton Annihilation at Darmstadt
(PANDA), an experiment that is one of
FAIRs four scientific pillars.
One potential solution was to scrap
PANDA, which an expert group had judged
to be the least internationally competitive pillar. But in September 2015, the
FAIR council decided to stump up the
missing cash rather than sacrifice the experiment, designed to study the structure
of protons and neutrons. The partner
countries agreed to find 158 million of
the missing funds (in 2005 prices) by the
end of this June and then settle the balance in 2019; that way, experiments could
start in 2022 and FAIR could be fully
operational in 2025.
After further negotiations, the FAIR
partners last week concluded that the necessary provision of funds is assured, says
FAIR Scientific Managing Director Boris
Sharkov in Darmstadt, Germany. He declined to say how much has actually been
guaranteed, and by whom, but admits that
neither Russia nor the third largest partner,
Indiadue to foot about 3.5% of the bill
has yet signed on the dotted line. Negotiations with both countries are progressing
very positively, Sharkov says.
But one of Indias two representatives
on the FAIR council, Bikash Sinha of the
Variable Energy Cyclotron Centre in Kolkata, says discussions have been acrimonious and that India and other countries
are not prepared to bear the costs of
strict German laws. FAIR may lose out to
other nuclear labs worldwide as a result,
he says, adding: Time is of the essence.
FAIR council member Vladimir Fortov of
the Russian Academy of Sciences did not
respond to calls from Science.
Physicist Angela Bracco of the University
of Milan in Italy, a former member of FAIRs
joint scientific committee, is confident that
digging of the synchrotron tunnel will start
next year. If the shortfall isnt made up, one
option under discussion is to use part of
FAIRs operational budget, she says. Bracco
also expects that some instruments will be
built or installed late to ensure others are
completed on time.
Wiedner says he doesnt know whether
technicalities or real problems are holding Russia back, but believes other FAIR
members would not have given the green
light without being rather confident that
the whole financing is available. He now
hopes that FAIRs membership can be expanded. Other countries have to have confidence that FAIR will really fly, he says. j


a few weeks of smooth sailing, the balloons

flight path during the long austral nights
and short days became as erratic as a zeropressure balloons: It dipped as much as
10 kilometers nightly from its cruising altitude of about 33 kilometers, possibly because
of a transient helium leak. And rather than
circling the Southern Ocean for 100 days
as intended, the balloon veered off over the
South Pacific after just one circumnavigation, having slipped out of the winter cyclone winds that circle Antarctica. Mother
Nature is in charge of our business, says
NASA Balloon Program Office Chief Debora
Fairbrother in Wallops Island, Virginia. She
truly has been exercising her rights. NASA
brought the balloon down in mountainous
western Peru and will seek to draw lessons
for future flights after retrieving it.
Still, the wayward balloon hauled in
some novel astrophysical data. It carried the
Compton Spectrometer and Imager (COSI),
a gamma ray telescope that aims primarily
to probe how elements are forged in supernovae. COSI observes gamma rays emitted by
radioactive nuclei in supernova debris, and
may be able to measure their polarization.
Such observations are hard to make from the
poles because of background gamma radiation from cosmic rays channeled toward the
poles by Earths magnetic field.
COSI, a lightweight instrument designed
for superballoon flight, was capable of sending data back in real time in case it couldnt
be recovered. Besides studying supernovae
and other gamma sources, the telescopes
long flight allowed it to detect a bright
gamma ray burst, says lead COSI investigator
Steven Boggs, an astrophysicist at the University of California, Berkeley. You have to be
in the right place at the right time to catch
[gamma ray bursts].
Stratospheric balloon observations have
limitations. For research that produces very
high volumes of data, such as studies of the
cosmic microwave background, we have
to store all the data onboard, says Shaul
Hanany, an astrophysicist at the University
of Minnesota, Twin Cities. Such research will
have to wait for superpressure balloons to
prove their reliabilitythat they and their
payloads wont be lost in the ocean.
Still, NASA has a full slate of superpressure balloon missions lined up in the coming years, studying phenomena ranging from
dark matter to cosmic rays. As more superpressure balloon projects produce useful
data, says Eliot Young, a planetary scientist at
the Southwest Research Institute in San Antonio, Texas, everyone is going to realize its a
great opportunity. j


Massive helium fields found

in rift zone of Tanzania
Commercial development could struggle in a stabilizing
market for the worlds best coolant
By Eric Hand

n 2013, Tom Abraham-James and Josh

Bluett, geologists and mates from
Australia, were on a holiday road trip
to Serengeti National Park in Tanzania. Like any good mineral explorers,
they kept a copy of Industrial Minerals in Tanzania: An Investors Guide in
their Land Cruiser. Several of its pages
described hot springs in the East African
Rift that seeped a seemingly boring, inert
gas. It was mostly nitrogen, the guide said,
with between 4.4% and 18.2% helium by
volume. Thats 10 to 100 times more helium than is typically found in natural gas,
the usual source. We thought there was a
discrepancy with the decimal places in the
article, Abraham-James says.
But after the researchers traced the data
to their sourcegeological surveys from
the 1950sthey began snatching up mineral leases. By 2015, they had founded Helium One, based in Bergen, Norway. And
last week, they disclosed details about
their claims to three massive fields of
heliuma nonrenewable resource and precious commodity not just for purveyors of
party balloons, but for scientists and medical technicians who need to keep things
really cold. (Liquid helium, with a boiling
point of 4 K, is the worlds best coolant.)
The firms best-characterized reservoir
contains 1.5 billion cubic meters of helium,
enough to satisfy world demand for 7 years.
Two other nearby fields could contain much
more. But the announcement is not just
about size; unusually, Helium One plans
to find and extract helium as a standalone
product, rather than by sieving it from natural gas. This is the first time that anyone
has prospected for helium in a deliberate
way, says Chris Ballentine, a geochemist at
the University of Oxford in the United Kingdom who presented his analysis of Helium
Ones fields last week at the Goldschmidt
Conference in Yokohama, Japan.
But prospectors have looked at Tanzania before, and experts question whether
Helium Ones fields are really worth developing in a relatively calm helium market
that has recently moved from shortage to


surplus. Now that theres a surplus, were

skeptical as to how well its going to work
economically, says John Campbell, a consultant at Intelligas in Dedham, Massachusetts. That could change when the rest of a
strategic U.S. helium reserve is auctioned
off a few years from now. But Campbell
adds that although large for a single field,
the discovery is small in the grand scheme
of things. The most recent assessment of
worldwide helium resources, prepared
by the U.S. Bureau of Land Management
(BLM) in 2006, estimated that the world

Pumped up
Rich fields in Tanzania could add to worldwide
supplies of helium, which is usually harvested as a
byproduct of natural gas production. But production
from other countries could crowd the newcomers out
of the market for the near future.





2015 helium production

(million cubic meters)

has 52 billion cubic meters of proven,

probable, or possible reserves.
Fresh helium comes from the decay
of radioactive uranium and thorium in
Earths crust. Over time, those elements
emit a particleshelium nucleiwhich
pick up electrons and end up lodged in
mineral defects as solitary helium atoms.
Old rocks have more time to accumulate
atoms. So the ideal source rocks are granites like those in the 3-billion-year-old Tanzania craton, an island of ancient crust in
Africa. The Tanzanian helium fields have
a second key ingredient: a source of
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS



Downloaded from on July 8, 2016

NASAs latest superpressure balloon and its gamma

ray telescope prepare to take off from New Zealand.


At Lake Balangida in Tanzanias rift zone, volcanoes are near enough for their heat to liberate helium from
underground rocks but far enough away not to dilute it with their own gasesperfect conditions for a high-yield
helium field, researchers say.



next step: a
ray biohybrid
Light-controlled cardiac
muscle cells guide
swimming of raybot
By Elizabeth Pennisi

evin Kit Parker wants to build a human heart. His young daughter loves
the New England Aquarium in Boston.
On p. 158, fathers and daughters obsessions have combined in an unlikely
creation: a nickel-sized articial stingray whose swimming is guided by light and
powered by rat heart muscle cells.
Incorporating advances in engineering,
cell culture, genetics, and biomechanics, the
living robot is clearly a technical tour de
force, says Adam Summers, an integrative
biologist at the University of Washington,
Seattle. And some think that by melding
cells and artificial materials into a pulsating
structure, the device brings Parkers dream
of engineering a human heart a step closer.
One can imagine that one day we can use
this technology to rebuild parts of the human
body, says Kedi Xu, a neural engineer at Zhejiang University in Hangzhou, China.
Parker, an applied physicist at Harvard
University, made his first foray into robotics 5 years ago after he was captivated by the
jellyfish during a visit to the aquarium. The
creatures rhythmic pumping reminded him
of the beating heart. His team had already
gotten heart muscle cells to grow into thin
films on silicone, and he wondered whether
he could put the cells to work by incorporating them into a jellyfishlike pump.
The result was a medusoida simple
artificial creature composed of heart muscle
cells overlaid on a sheet of silicone molded
into a shallow cup rimmed with flaps. A bath
of salt-sugar solution sustained the cells,
and tiny jolts of electricity made the cells
contract, changing the shape of the silicone
cup so that the jellyfish expelled liquid, propelling it through its bath. For me this was
just a training exercise, Parker recalls. Im
trying to get better and better at building
muscular pumps.
His new effort was also inspired by a visit
to the aquarium. When his daughter touched
a stingray in the petting tank, it flicked one SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016

2005 and 2015, reaching $3.75 per cubic

meter, driven by growing demand from
medical imaging facilities. And cheap sales
from the U.S. helium reserve, which for
many years kept crude helium prices low,
are about to dry up. In 2013, Congress altered the law so that the reserve would be
auctioned off closer to market prices. The
reserve could be exhausted by 2019, says
Robert Jolley, the BLM field office manager
in Amarillo, Texas.
Still, Garvey says, prices have plateaued
in the past year because of new helium
production coming online in places like
Qatar just as demand has slackened. Welders have shifted to using argon rather than
helium, and MRI facilities and scientists
have invested in machines that recycle
helium. The problem is, we have plenty of
helium right now, she says.
And other suppliers are eyeing the imminent end of the U.S. reserve. Russias Gazprom, which owns vast stores in the Far
East, plans to boost helium production in
2023, Campbell says. For projects like the
ones in Tanzania to have long-term viability, Campbell says, new demand pressure
must arrive. One long shot: fusion power,
which would need helium to cool its superconducting magnets. If fusion energy ever
were to be developed and become a significant power provider, that would use all the
helium we could muster. j


heatin this case the volcanic systems of

the East African Riftthat can cook the
rocks and release the helium into traps
consisting of porous rock capped by an
impervious layer.
Ballentine reported helium concentrations of between 2.5% and 4.2% at the
Ruksa field, where the gas seems to be
trapped in sandstone reservoirs 0.5 to
2.5 kilometers down. Two other fields, Balangida and Eyasi, contain much higher helium concentrations, up to 10.5%, although
their volumes have not been explored as
fully. If Helium One can raise $40 million, it plans to begin drilling in the second quarter of 2017, Abraham-James says.
He says the Tanzanian fields can provide a
standalone helium supply that is independent of the price of natural gas. You can
turn up and turn down helium production
according to market demand, he says.
Maura Garvey, a consultant for Intelligas, is skeptical. She consulted on a 1998
assessment of a nearby helium field in Tanzania by a different company. We killed
it, she says. We couldnt justify the price
of their helium. There was no local market, roads were poor, and prices at the time
couldnt justify the capital cost of liquefying helium for export, she says.
Those conditions have improved since
the late 1990s, she acknowledges. Prices
for crude helium nearly doubled between


side of its body and veered away. Maybe

there is something similar with how the
stingray changes direction and heart flow,
Parker thought. So he decided to move up the
tree of life, from jellyfish to rays, and increase
the complexity of his teams biohybrids.
Again thanks to his daughter, he envisioned a simple way to control the new robot: light. When she was a toddler, Parker
would guide her down the sidewalk by shining his laser pointer on the ground and having her stomp on the light. Perhaps his team
could do something similar in an artificial
ray by making the muscle cells contract
in response to light. They turned to optogenetics, in which cells are genetically endowed with light-responsive molecules that
trigger signaling cascades. Parkers team had
no experience with the method, but with a
bravado befitting a lieutenant colonel who
fought in Afghanistan before and after joining
Harvards faculty, he decided to go ahead
cobbling together funding from the Army,
the National Institutes of Health, and others.
Parker asked a new postdoc, Sung-Jin Park,
to oversee the work, confidently predicting
the project would produce a paper featured
on the cover of Science. I thought it was
crazy impossible, Park recalls.
It took 4 years to fulfill Parkers prediction.
Park and others began by taking apart stingrays to learn how the muscles were arranged;
later, colleagues in another lab analyzed how
the muscles drive the fins in the synchronized
undulations that propel the ray. To mimic the
animals basic anatomy, Park experimented
with many soft-robot configurations, eventually settling on a multipronged gold skeleton
sandwiched between two silicone layers.
Animating each ray are about 200,000 heart
cells harvested from 2-day-old rat embryos
and placed on top of the silicone. The silicone bears a template consisting of an extra-

The rays move only

about 9 meters per hour
and turn slowlyquite
pathetic by real stingray standards. Even so,
putting it all together
is remarkable, says
Alexander Smits, a
at Princeton University. Fish, who helped
design larger manta
bots with silicone fins
flapped by electroniccontrolled rods and
cables, calls the rays a
cellular protein, fibronectin, which guides the
major leap forward in terms of robotics.
growth of the cells into a radiating pattern
He adds, were getting to the point where
similar to the muscles in the real ray. Getting
there really is a fusion between biology
this architecture right was critical to making
and engineering.
cardiac cells do the work of skeletal muscles,
Theres a long way to go, however. Livesays Andrew McCulloch, a bioengineer at the
muscle robots work only in nutrient-filled
University of California, San Diego, who was
solutions warmed to a rats body temperanot involved with the work.
ture; keeping them going in a more natural
But Parkers group didnt follow the rays
environment will be a challenge, Fish says.
muscle structure exactly. They took a shortAnd its unclear whether the approach
cut, says Frank Fish, a biomechanist at West
will lead to practical robots or to Park and
Chester University in Pennsylvania. Real rays
Parkers true interest: a bioartificial heart.
have two sets of muscles
Simply putting down a
within each pectoral
second layer of muscle
fin, pulling in opposite
cells is daunting, nevdirections to move a fin
ermind recapturing the
down, then up. The ray
full complexity of the
robot has just one set
heart or a complete aniof muscles, which bend
mal. The ray robot is
the fins downward; the
not really particularly
spring action of the gold
related to anything in
skeleton pulls the fins
the heart, especially beback up.
cause the heart muscle
Infected with a virus
cells are being used in
that delivers the gene
a fairly unnatural way,
optosays Denis Buxton, a
genetic molecular switch,
program officer at the
the modified cardiac
Kevin Kit Parker and his daughter compare
National Heart, Lung,
cells twitch when blue
a ray robot on a slide with a skate.
and Blood Institute in
light shines on them.
Bethesda, Maryland.
But translating that effect into coherent
But to Parker and others that unnatural
motion took months of tweaking; simway is very informative. The heart is a
ply getting a robot ray to move forward
hollow muscle, explains Simon Hoerstrup,
when light stimulated the front of its fin
a cardiovascular surgeon at the University
took Park 200 tries. Ultimately, he built
of Zurich Institute for Regenerative Medi100 more robots and showed they could navcine in Switzerland. Many of the features
igate underwater obstacle courses. To negoyou see in this ray, you find in the heart.
tiate turns, Park guides a ray with two light
Parker regards his teams miniature ray
sources, one pointed at each fin. Changing
robot as a piece of art as well as technology:
the frequency of the light slows or speeds up
Everyone is going to see something difthe contraction rate; by making one side beat
ferent in it, he says. Im looking at it and
faster than the other, he steers the robot left
Im trying to understand the heartand
or right.
impress my 7-year-old daughter. j


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

Part machine and part living

cells, this artificial stingray
can be maneuvered through
obstacle courses.



In Canada, peer-review changes stir outrage

Reforms by nations lead biomedical funder result in shoddy reviews, critics say


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016

established scientists and were skeptical of

online reviews. After the first Foundation
grant competition in 2015, success rates and
award levels for both female and younger
researchers were lower than for male, more
elderly counterparts, says geneticist Janet
Rossant of The Hospital for Sick Children
and the University of Toronto in Canada,
and a former member of CIHRs governing council. The criteria, I think, became
very much skewed not toward outstanding
science, but outstanding leadership, says
Rossant, a winner in the competition.
Shifts in CIHRs spending resulting from
the program restructuring added to the disadvantage for early career researchers, says


Conservative Prime Minister Stephen Harper

in 2013, rebuffed the request.
ne day, historians might call it the
Worries about the new peer-review sysPeer Uprising.
tem crystallized earlier this year, after CIHR
Nearly 1000 Canadian researchconducted its first Project grant competiers are demanding that the governtion. Tight funding had caused CIHR to
ment immediately reverse radical
cancel two previous competitions, so the
changes that the nations main biomajority of this countrys health researchmedical research funder has made to its
ers applied to the latest competition, notes
system for awarding research grants. In
the 27 June letter, which was drafted by
particular, the researchers want the CanaJim Woodgett, director of research at the
dian Institutes of Health Research (CIHR)
Lunenfeld-Tanenbaum Research Institute
to scrap its new online system for evaluatat Torontos Mount Sinai Hospital. Despite
ing grant proposals.
warnings, CIHR scrambled to find sufficient
The new system shows deep flaws and erroreviewers, at least some of whom were inexneous assumptions and has led to shoddy reperienced or unqualified.
view practices, the critics charge in a
The shortage of reviewers forced
27 June open letter to Jane
the agency to rely on a matching alPhilpott, Canadas minister of
gorithm that assigned applications
health. There is little hope that
to reviewers based on keywords.
the best ideas and projects will be
Some were asked to assess applicafunded. We posit this represents
tions in disciplines that they knew
a fundamental failure of CIHRs
nothing about, Woodgett says. Reprimary mandate.
viewers, who often had just days to
The letter represents the latrespond, sometimes didnt bother
est salvo against a controversial
to provide written reviews or particreform effort launched roughly
ipate in the online discussions, and
4 years ago by Alain Beaudet, presiapplicants were given no chance
dent of CIHR, which spends about
to submit revisions in response to
$790 million annually on health
feedback. The new system may be
research. In a move designed to
more cost-efficient, Woodgett says,
create more stable funding for the
but it represents the abandonnations elite researchers and rement of scientists evaluatduce time spent chasing grants, the
ing other scientists in front of
agency essentially obliterated its
other scientists.
existing grant programs, replacing
Critics say CIHR needs to beat
them with just two flavors: Founa hasty retreat from its reforms.
dation scheme grants that provide
Jim Woodgett of Toronto, Canadas Mount Sinai Hospital drafted a protest
They want Philpott to immediately
$38,800 to $1.16 million annually to
letter signed by nearly 1000 researchers.
meet with a representative group
established researchers for openof scientists to iron out concerns,
ended research and Project scheme grants
Michael Hendricks, co-founder of the Assoand for CIHR to revert to traditional peer
of $38,800 to $581,000 annually for up to
ciation of Canadian Early Career Health Rereview until the new methodology has been
5 years for more focused projects.
searchers and assistant professor of biology
assessed by international experts. The governCIHR also remade its grantsmaking proat McGill University in Montreal, Canada.
ment should also convene a national summit
cess. Under the old system, applications
The shifts took almost half of the open,
on health research.
were distributed among 53 discipline-based
operating grants funds off the table for new
Beaudet declined Sciences request for
expert panels, which met face-to-face to
investigators, he says. Were looking at
comment. But he has posited that the critirank applications. Now, researchers subprobably over a 30% loss, in dollars, in fundcism of CIHR is linked to funding struggles,
mit applications to a central pool; each aping support.
and will dissipate if Prime Minister Justin
plication is assigned to four reviewers, who
As a result of such concerns, earlier this
Trudeaus new Liberal government boosts
are supposed to participate in a virtual, ayear the U15 group of research-intensive Caspending for the health agency. Its unlikely,
synchronous electronic discussion through
nadian universities asked CIHR to put a temhowever, that money alone would quell
an internet chat room. Proponents argued
porary moratorium on further changes to
the uprising over the perceived threat to
the changes would boost efficiency, reresearch funding programs and to convene a
the time-honored tradition of face-to-face
duce reviewer fatigue, and promote multinational summit to discuss the changes. But
peer review. j
disciplinary projects. Critics, however, worBeaudet, who in 2013 was appointed to a secried that the funding changes would favor
ond 5-year term as CIHR president by former
Wayne Kondro is a writer based in Ottawa.
By Wayne Kondro

Kala gave birth to the first

brown howler monkey born in Tijuca
Forest in more than a century.

A team of ecologists is recreating
a living rainforest in the heart of the Olympic city


By Herton Escobar, in Rio de Janeiro, Brazil

een from the Corcovado, the

710-meter-high gneiss mountain
topped by the iconic Christ the
Redeemer statue, Tijuca National
Park rolls out below like a lush
green carpet, edged with Rio de
Janeiros urban sprawl. As the city
scrambles to finish preparations
for the Summer Olympics next

month, scientists here are playing a longer

game. They are embarked on a decadeslong effort to transform Tijucaone of the
biggest urban forests in the worldfrom
a verdant forest park inhabited mainly by
hikers and stray dogs into something more
akin to a genuine rainforest. People have
this nave impression that the forest is full
of animals, says Fernando Fernandez, an


ecologist at the Federal University of Rio de

Janeiro (UFRJ). Our goal is to fill this forest with life again.
Decades ago, an expanding Rio de Janeiro
surrounded Tijuca, severing its 4000 hectares from other forest tracts. Jaguars were
long gone, but now hunters and loggers
extirpated the monkeys, rodents, and other
mammals living in the forest island, setting
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016



8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016

in motion profound ecological changes.

change the course of evolution, Fernandez
destruction started early. Around the time
Trees that rely on animals to disperse their
says. So, if we bring those species back, can
Darwin was exploring Tijuca, vast swaths of
large, hard seeds gave way to less dense,
we put evolution back on track?
forest around what was then Brazils impefaster-growing trees whose seeds are disrial capital were being razed and replaced
persed by wind or water. The forests carbon
ONE OF THE LAST to record Tijuca as a living
with coffee plantations. Within a few years,
storage capacity shrank. Every aspect of an
forest, resounding with the guttural calls
Tijuca itself was mostly leveled. As a result of
empty forest is a disaster, Fernandez says.
of brown howler monkeys, was Charles
the deforestation, the city started running
His team has embarked on a pioneering
Darwin himself. During his epochal voyage
low on water and temperatures soared,
effort to restore Tijucas original fauna. Six
on the Beagle, Darwin stopped here and exprompting authorities to launch a major reyears ago, they began releasing red-rumped
plored the nearby forest. As he noted in his
forestation effort in the 1860s.
agoutis (Dasyprocta leporina), a rodent
diary, on 4 June 1832 a young hunter led
Laborers replanted hundreds of thouthat is an important seed disperser, fitted
him to a spot in the forest where he had
sands of native and exotic trees. But the
with radio collars. The refaunation effort
shot two large bearded monkeys the day
fauna did not bounce back. Fruits are the
entered a new phase last September, when
before. One of the dead howlers still clung
bribe that plants offer to animals to disperse
the team released four brown howler monto a tree with its prehensile tail.
their seeds. So if you see fruits rotting on the
keys (Alouatta guariba) in Tijuca, and more
Tijuca is a fragment of Brazils Atlantic
forest floor, its an empty forest, Fernandez
will follow this year.
Forest, which once covered most of the counsays. His team is seeking to complete Tijucas
The refaunation stands out in restotrys Atlantic seaboard but now is a fragdrawn-out recovery by adding animals. Esration ecology for its attempt to heal
mented 12.5% of its former self. Near Rio, the
sentially, what we are doing is a continuation
an entire ecosystem, conservaof that early reforestation process,
tionists say. In contrast to many
says biologist Ernesto Castro, park
manager at Tijuca. The forest is so
Repopulating an empty forest
Fernandezs team isnt trying to save a
isolated that the only way for these
In the 19th century, vast swaths of Tijuca were razed. Bringing back
particular endangered species. The
animals to return is to bring them
native mammals is an extension of reforestation efforts.
coolest thing is that they are working
back ourselves.
with very common species[ones]
In 2008, one of Castros
that are not threatened, but are very
predecessors, Ivandy Castroimportant for the environment, says
Astor, suggested the refauMauro Galetti, a conservation bionation project to Fernandez.
logist at So Paulo State University,
Fernandez saw a chance not
Rio Claro, in Brazil. Regrowing a
only to rebuild a beloved
forest is relatively easy. Rebuilding
park, but also to conduct a
an ecosystem is a lot harder.
large-scale ecological experiThere is no formula for restoring
ment. He brought back agoutis
a tropical forest, Fernandez says, so
first, he says, because they are the
he and his colleagues are proceedbest seed dispersers in the world.
ing cautiously, introducing species
More than 20 tree species at the
one by one and monitoring how
park depend heavily on these roTijuca National Park
Rio de Janeiro urban area
each one fares and how it affects the
dents to gather their seeds and bury
forest. Every time you lose species,
them in caches all over the forest.
Area where
you lose ecological interactions. And
Those trees were being held hosChrist the Sugarloaf
animals were
when you change the ecology, you
tage, reproductively, by the absence
Redeemer Mountain


The brown howlers radio collars are broken or lost, so Lusa Genes spends many hours tracking down the monkeys on weekly expeditions to monitor their behavior.


monitor the monkeys now is by naked

eye. That takes a lot of patience: The monkeys dont howl often, and spend most of
their time quietly chewing leaves high in
the canopy.
Lusa Genes and Tomaz Cezimbra, two of
Fernandezs graduate students, are responsible for tracking the primates. They note
the howlers behaviors and mark with pink
tape the trees they sleep in or feed on for
later identification. Genes also scans the

the canopy, and there Kala wasin a place

he and Genes had passed several times
beforecollecting leaves with her hands
while gripping a branch with her tail.
Genes and Cezimbra were happy to find
Kala, but they worry about Juca, who was
not with her. It was the first time they had
failed to spot the baby. Its a bad sign,
Fernandez says. This is not a success story
IN LATE MAY, a visitor in Tijuca heard a
yet, he says. It could fail at any moment. As
baby wailing in the forest, and sounded the
Science went to press, Juca was still missing.
alarm. Authorities dispatched
Theres a chance it could be with
a ranger to find the lost infant.
the other female, Genes says.
To everyones relief, the ranger
But it could be dead.
quickly tracked down the source
To boost the odds of establishof the plaintive ululations: a
ing a viable howler population,
brown howler, maybe 6 months
Fernandezs team plans to release
old, bawling as its mother, Kala,
10 more by years end. Provided
munched leaves on a nearby
that goes well, another canditree. The cries had sounded like
date for reintroduction is the ena parents nightmare, but they
dangered golden lion tamarin
represented a triumph: the first
(Leontopithecus rosalia), a flaghowler baby born at Tijuca in
ship species of Brazilian biomore than a century.
diversity conservation, which
Of the four adult howlers reis also a seed disperser. For the
leased last fall, one pair came
tamarins to move in, however, anfrom a private breeder, and the
other small primate, the common
other from a federal facility for
marmoset (Callithrix jacchus),
seized animals. They had lived
must be evicted first. It is one of
for 8 months in adjacent cages
several exotic species that have
at the Rio de Janeiro Primatoprospered in the park after the
logy Center getting accustomed
original inhabitants disappeared.
to each other and to a leaf-based
Other candidates for reintroducdiet, similar to what they would
tion are the maned three-toed
have in the wild. After that, they
sloth (Bradypus torquatus)a
spent 20 days in a pen in Tijuca
voracious leaf eater and nutrito adjust to the environment.
ent recyclerand the kinkajou
Fernandezs team was anx(Potos flavus), a seed disperser
ious to see what would happen
and pollinator.
once they opened the pen, as
Larger mammals of Brazils
the monkeys had never climbed
Atlantic Forest, such as the tapir
a tree before, says Marcelo
or the jaguar, will never return.
Rheingantz, a biologist at UFRJ.
Tijuca is too confined for them;
But as soon as we opened the
prey is too scarce and conflicts
door, they all climbed up immewould arise with surrounddiately. Three mostly stay up in
ing communities. Maybe some
the trees. One male, the youngest
smaller cats, such as the oncilla
of the four, settled in public areas
(Leopardus tigrinus), could evenof the park and began taking food
tually be reintroduced after prey
from visitors and sunbathing
return to sustainable levels, Feron the park road. The scientists
Agoutis were brought back first because they are supreme seed dispersers.
nandez says. The oncilla would
brought him back into captivity,
Lusa Genes (bottom) holds a hard rind gnawed by one of the rodents.
be good competition for the feral
for his own safetyand to prodogs, now the top predators in
tect humans who might provoke him. He
forest floor for monkey feces and insects.
Tijuca and a threat to the agoutis and other
wasnt fulfilling any ecological role there,
She is studying how the howlers influence
smaller critters.
Rheingantz says. The other male also began
the ecology of the forests dung beetles,
But the cats wouldnt arrive for years: a
interacting with people, and was moved to
some species of which feed exclusively on
late thread in the slow, patient process of rea more remote location. One of the two falarge primate excrement and are important
pairing Tijucas frayed fabric. We will start
thered Kalas baby, named Juca.
for secondary seed dispersal.
with the easiest animals, Fernandez says,
Keeping tabs on the monkeys isnt easy.
Kala is the easiest to monitor, as she
and try to reconstruct the ecosystem little
All four were released with GPS tracktends to stay near the pen area. Still, on a
by little. j
ing collars, but three collars soon malfoggy day in early June, it took 7 hours to
functioned and Kala lost hers after
find her. Cezimbra spotted movement beHerton Escobar is a writer in
6 months. So the only way to locate and
hind a clump of branches 20 meters up in
So Paulo, Brazil.

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016


of the agoutis, Galetti says. A forest without

animals is like a stage without actors.
By now at least 35 red-rumped agoutis
roam Tijuca. And this year they have an accomplice: the first howler monkeys, which
contribute to seed dispersal and nutrient
recycling as they move around the canopy,
eating and defecating.

Downloaded from on July 8, 2016

Marcia McNutt on the grounds

of the National Academy of
Sciences building, with Albert
Einstein looking on.

Meet Marcia McNutt, scientist, administrator, editor,

and now National Academy of Sciences president
By Ellen Ruppel Shell


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS






8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

hen geophysicist Marcia

in Palo Alto, California, partly for its promthing about tacking into the wind. Eureka! I
McNutt took over as director
ise of closer contact with faculty but not
tried one thing after another, and somehow
of the U.S. Geological Survey
this sort of contact. No one had ever told
learned the difference between jibing and
(USGS) in 2009 as part of the
me I couldnt do something, she told me.
coming about. Eventually, I made it back to
new Obama administrations
The one thought in my mind was, Ill
shore. After that, I was an experimentalist.
dream team of scientistshow him!
That experimental bent led her to physadministrators, her first priMcNutt went on to ace all four of Wrights
ics and mathematics as an undergraduate
ority was to reorganize the
required classes and graduate summa
and then to the geosciences, particularly the
agency to respond to realcum laude with a degree inof course
then-nascent theory of plate tectonics. It was
world problems. But USGS scientists, many
physics. Years later, as a chaired professor
a field wide open to a young scientist eager
of whom had been with the agency for deof geophysics at the Massachusetts Institute
to make her mark: Major expeditions were
cades, were known for their resistance to
of Technology (MIT) in Cambridge, McNutt
often fully staffed and even led by graduate
change, so McNutt devised a remarkable
declined an invitation to pen a testimonial
students, as many senior scientists were hesstrategy. She could not fire department
about Wright in honor of his retirement.
itant to embrace the new paradigm.
heads, but she could assign them
While in graduate school at the
to a regional office outside their
Scripps Institution of Oceanobeloved Menlo Park, California,
graphy in San Diego, California,
and the post she offered was
McNutt relished the opportunity
her home town, Minneapolis,
to follow her research wherever
Minnesota. One by one, McNutt
it took her, especially when it led
recalls, department heads reupstream from received wisdom.
tired or quit, leaving her free to
I wanted to go places, see things
set a new direction.
personally, collect data, and
We were living in geologic
revise on the fly, she says. But
time, so Marcia took some
rather than focus on what other
getting used to, says Bill
researchers were studyingthe
Werkheiser, now deputy direcboundaries between plates,
tor at USGS in Reston, Virginia,
where most of the action seemed
who was McNutts associate dito be taking place (earthquakes,
rector at the time. She made devolcanoes, mountain buildcisions very quickly we knew
ing, ocean trench creation)
we had to change, and she made
she turned her attention to the
it happen. But she was always
plate interiors, and to a mystery
clear, you knew where you stood,
that had thwarted other scienand she was fiercely loyal to us.
McNutt is an expert barrel racer, an event where speed and control are key.
tists: why so much volcanic acThat mix of decisiveness, hutivity appeared to be happening
manity, and negotiating skill served McNutt
He wouldnt want to read what I had to
so far from the plate boundaries.
well both as a researcher and an administrasay, she says firmly.
Sean Solomon, currently director of the
tor: In addition to USGS, she was the first
Wright wasnt the last person to underLamont-Doherty Earth Observatory at Cofemale president and CEO of the Monterey
estimate Marcia McNutt, or the only one
lumbia University, was instrumental in reBay Aquarium Research Institute (MBARI)
to regret it. I bow my head to Marcia,
cruiting McNutt to MIT in 1982, when he
in Moss Landing, California, and, from 2013
says MIT physical oceanographer Paola
was on the faculty there. He recalls being imuntil last week, the first female editor-inMalanotte-Rizzoli. She has a spine of iron.
pressed by the young scientists tenacity and
chief of Science. This month, she became the
drive, and by her ability to speak convincfirst female president of the National AcadMCNUTT GREW UP IN MINNEAPOLIS and spent
ingly to both scientific and lay audiences. All
emy of Sciences (NAS), the governments preher childhood summers in a lakeside cabin,
of us knew who the great graduate students
mier science advisory organization.
where reading was the default rainy day acwere, and she was clearly among them. She
None of this has been easy. Indeed, starttivity. She spent many a morning at a nearby
was audacious, quite willing to go where
ing with college, hurdling obstacles has
farm, mucking out the stables in exchange
others wouldnt.
been a constant in her life. In the fall of 1970,
for a chance to ride. (She later became an
Scientists had long noted that the ocean
William H. Wright, a professor of physics at
expert barrel racer, a rodeo event that enfloor deepens with increasing distance from
Colorado College in Colorado Springs, ushertails galloping around a tight cloverleaf of
the ocean ridges, where new crust is creed freshman Marcia McNutt into his office
barrels; speed rather than finesse is paraated. But an area beneath French Polynesia
with what she recalls as this observation:
mount.) But perhaps her most vivid memory
posed a puzzle: Why was that swath of ocean
You are here because you must have said
of those summers involves an old Sunfish
thousands of miles from a plate boundary
something silly on your application about
that an uncle dropped off for the familys use.
so shallow and its floor riddled with volcabeing a physics major. Ive seen girls come
Too impatient to wait for an adult to teach
noes? McNutt plunged in. Using sonar sigand go in this department, but Ive yet to
her the fine points of navigation, she jumped
nals to map the seafloor topography, she
see one graduate. For perhaps the first
into the boat and set sail. I knew nothing,
discovered what she described as a supertime in her young life, McNutt was struck
just pushed off from shore and in no time
swell, a broad region of unusually shallow
speechless. Class valedictorian at the allwas out in the middle of the lake, she reocean floor buoyed by hot rock welling up
girls Northrop Collegiate School in Minnecalls. I could see our cabin, but I had no idea
from the mantle. In a groundbreaking paper
apolis and with perfect SAT scores, she had
how to get back. Then I remembered reading
in Science, McNutt concluded that the rocks
chosen Colorado over Stanford University
a Nancy Drew story that mentioned someexcess heat and extremely low viscosity had

role model. None of us thought we could do

what Marcia didshe played at a level well
beyond [the level to] which the rest of us
were headed.
MCNUTTS AMBITIONS went beyond aca-

demia. As a scientist, working in a lab, publishing papers that only a few specialists in
the field really cared about, it felt to me like
being trapped in a box canyon, she says. She
determined that what she calls her highest and best use was not doing science, but
enabling other scientists to do theirs. So she
did what almost no one else in her situation
would do: In 1997 she left a tenured position at MIT, packed up her daughters, their
nanny, Ann, and Anns daughter, and moved
to Salinas, California, (salad bowl of the
world) to run MBARI. Her leaving was a
terrible loss for MIT, Malanotte-Rizzoli says.
She wanted a new experience, and she deserved that. But geophysics suffered.

of public (and funders) concern, such as

protecting the oceans from acidification and
algae blooms and understanding the role
the oceans play in climate change. Under
her leadership, MBARI built a chemical sensor laboratory to detect ocean pollutants in
real time, and grew an autonomous underwater vehicle program to make sample collection safer and more efficient. And, somehow, despite the economic downturn, it
doubled its staff.
While at MBARI, McNutt also served
as president of the American Geophysical
Union (2000 to 2002). And yet, once again,
she wanted greater influence. So when Sean
Solomon met with her in 2008, this time
as chair of an NAS committee convened to
recommend a new USGS head, McNutt listened. Joining the Obama administrations
dream team of scientific administrators,
she decided, would be her next highest and
best use.

Presenting at the fall 1992 American Geophysical Union meeting, when McNutt was at MIT (left); heading a task force following the Deepwater Horizon disaster in 2010,
as U.S. Geological Survey director (center); on the Monterey Bay Aquarium Research Institutes flagship vessel in 1999, when she was the institutes CEO (right).

her graduate students, once telling off another senior scientist for showing disrespect
to a female member of her team.
For most of that time she was bringing
up three daughtersMeredith, Ashley, and
Danaas a single parent. Their biological
father died suddenly when the youngest,
identical twins, were only 2 years old, and
the oldest not yet in school. (In 1996, McNutt
married Ian Young, an MBARI ship captain.)
That Marcia managed to hold everything
together under those crushingly difficult circumstances seemed to us amazing, Fischer
says. At a time when not all that many
women were succeeding in science, she
made it normal for women to succeed.
Geophysicist Carolyn Ruppel, who was
one of McNutts MIT advisees at the time,
has a more nuanced take: Shes an amazing person, an amazing scientist, and has
made significant contributions in areas that
were difficult to navigate. But she was not a

MBARI, funded by the David and Lucile

Packard Foundation, was a relatively new
venture with a lofty mission: to apply cuttingedge technology to inform and shape the future of the worlds oceans. McNutt took the
reins shortly after Founding Director David
Packard died, and, according to aquarium
Executive Director Julie Packard, she quickly
bridged the leadership gap. One of my fathers most deeply held principles was to
invest in people and give them the space to
pursue their ideas, Packard says. Marcia
built and expanded on that vision. She took
a big risk leaving MIT where she was at the
top of her game to come to what was basically a startup operation. And we were very
lucky she did.
McNutt soon faced an unexpected challenge: The market plunge after 11 September
2001 sharply eroded the foundations assets.
With fewer resources at her disposal, she directed staff to dig even deeper into matters

McNutt wasnt alone in recognizing the

need to overhaul the structure of USGS.
USGS was full of ivory tower types each in
his or her own silothe seismic folks, geology folks, public health folks, says her thenboss David Hayes, who was deputy secretary
of the interior under both President Bill
Clinton and President Barack Obama and is
now on the faculty of Stanford Law School.
When Marcia arrived the agency was not,
to my view, living up to its potential to provide the science needed to help undergird
smart decision-making. She reorganized it
to align with todays science challenges
climate change, land useand she did it in
a remarkable way, with a sense of openness
and respect.
Her strategy of offering department
heads reassignment to the Minneapolis regional office could have taken a toll on morale, but USGS Deputy Director Werkheiser
recalls that Marcia was a great boss, and SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


allowed the volcanism to readily pierce the

lithospheric plate, liberating fully 30% of the
heat flux from all hot spots on Earth in that
patch of Pacific Ocean floor.
This and a number of other discoveries
brought McNutt many accolades, among
them the prestigious Macelwane Medal from
the American Geophysical Union. Karen
Fischer, now a geophysicist at Brown University, was one of McNutts graduate students
at the time. She recalls the awards ceremony
as a quintessentially Marcia experience.
She wore an Oscar[s]-style evening gown
and looked incredibly glamorous, Fischer
says. We were incredibly proud of her she
had succeeded in science on her own terms.
McNutt made a practice of setting her own
terms. She rode a red Honda 500 motorcycle
to the office, always wearing fashionable
footwear. She made more than a dozen ocean
expeditions, and spent months at a time in
Tibet and Tahiti. And she fought fiercely for

Downloaded from on July 8, 2016



MCNUTT LEFT USGS IN 2013, many assumed

to take over as head of the Scripps Institution of Oceanography. But she chose to remain in Washington, D.C., to take the helm
at Science, where, among other things,
she presided over the founding of Science
Advances, an open-access, peer-reviewed
journal that reflects her concern with maintaining scientific integrity in an increasingly cutthroat publishing environment.
At Science, the paradigm is changing,
she says. Were talking about asking au-

McNutt and family. Marcel Hoffman, her first husband, died suddenly in 1988, when their three daughters
were not yet school-age. She married Ian Young (right), a research ship captain, in 1996.

thors, Is this hypothesis testing or exploratory? An exploratory study explores new

questions rather than tests an existing
hypothesis. But scientists have felt that
they had to disguise an exploratory study
as hypothesis testing, and that is totally
dishonest. I have no problem with true exploratory science. That is what I did most
of my career. But it is important that scientists call it as such and not try to pass
it off as something else. If the result is
important and exciting, we want to publish exploratory studies, but at the same
time make clear that they are generally
statistically underpowered, and need to
be reproduced.
McNutt put her stamp on the editorial
page of Science with some 60 editorials in
3 years as editor-in-chief. But she admits
that she took a wrong turn on the Keystone
Pipeline, a proposed route for oil produced
from Canadas oil sandsa project she regrets having publicly endorsed. I would do
things differently now, she says. I should
have said that I would support Keystone if
this happens, if being changing the process for extraction to make it cleaner, taxing the pipeline so that there is no decrease
in the cost of oil, and imposing environmental scrutiny.
McNutt also regrets the sexism scandal
that rocked Science beginning in July 2014,
when the journal published a cover photo
featuring a pair of transgender women
in platform shoes and skin-tight dresses
with their heads cropped out of the shot.
Though McNutt publicly vowed to strive
to do much better, that faux pas was followed by another: a Science columnist who
advised a female scientist to put up with
a male superior sneaking glimpses down
her shirt. Had I known, I would not have


run that column, she says, adding that the

editor involved no longer works for Science.
Women have to decide for themselves
what path to take in a situation like that,
find a resolution that allows her to go on
with her career and allows her to feel okay.
Now, McNutt will have an even higher
profile as head of NAS. My hope is that she
will be an outspoken public face for science,
with a focus and emphasis on evidence-based
decision-making, says Diane Griffin, professor of molecular microbiology and immunology at Johns Hopkins Bloomberg School
of Public Health in Baltimore, Maryland,
and vice president of NAS.
McNutt is more than ready to embrace
the challenge. She sees her job at NAS as
improving reproducibility and ethics in
science, promoting women in science,
and guiding the public conversation toward an understanding of science not as
a bloodless series of facts, but as a structured approach to elucidating the laws of
nature. The academy has the job of providing scientific advice to government,
and thats a role that has never been more
vital, she says. Its not the role of the
academy to say what the policies should
be, but it is the role of science to project
the consequences. Advice from the academy could be transformational to help the
nationand the worlddo the right thing.
Once again, and perhaps not for the last
time, it seems that McNutt has found her
highest and best use. j
Ellen Ruppel Shell, professor and codirector in the graduate program in science
journalism at Boston University, is author
of Cheap: The High Cost of Discount
Culture. Colin Norman, who was Sciences
news editor until 2013, edited this article.
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

I cant think of anyone who had serious issues with her.

She needed that support to handle a series of jolting disasters in her first 6 months
at USGS: major earthquakes in Haiti and
Chile, a water crisis in California, an invasion of Asian carp in the Great Lakes, and
a volcano in Iceland that disrupted transAtlantic air travel for nearly 10 days. And
then, on 20 April 2010, came what McNutt
calls her Omaha Beachthe explosion of
BPs Deepwater Horizon oil rig, which killed
11 workers, injured 17, and over a period of
87 days released 4.9 million barrels of crude
oil into the Gulf of Mexico. McNutt says she
and her colleagues were caught totally unawares, thanks in part to deceptive assurances from the oil industry that you dont
have to worry about oil escaping from
deep-sea wells.
Anyone who has spent more than 5 minutes with Marcia McNutt knows this: She will
patiently suffer fools but has zero tolerance
for deceit. She rushed to Houston, Texas, with
an overnight bag, expecting a short trip. She
ended up spending 4 months in a windowless 2-by-3-meter office at BP headquarters,
huddling late nights and early mornings
with scientists and engineers, calling every
expert she knew who might offer insight.
Soon after her arrival she was tapped to
lead the Flow Rate Technical Group charged
with gauging the volume of oil erupting
from BPs well, a highly contentious issue.
BP put the number first at 1000 and then at
5000 barrels a day. The groups estimate,
based on bits of high-definition video footage Congress had forced BP to share, was
far higher: as much as 60,000 barrels a
day. Othersin government, academia,
the presswere shooting from the hip,
McNutt says. But we had the data. Pushing through the bluster of what she called
BPs cowboy, get it done and go home attitude, McNutt announced the technical
teams findings to the world. Marcia was
pragmatic, she understood what needed to
be done to bring the stakeholders together,
Hayes says. And she showed a surprising
willingness to let it ripshe wrote some
emails she shouldnt have, believe me.

Energy-efficient chemical
separations p. 121 c


Downloaded from on July 8, 2016

Powering the expanding

Internet p. 124


Learning to move on land

Interdisciplinary research suggests that early
four-limbed vertebrates relied on their tails
By John A. Nyakatura

ow did early four-limbed vertebrates,

or stem tetrapods, move on land? On
page 154 of this issue, McInroe and
co-workers bring together expertise
from several fieldsincluding biomechanical analysis of a modern analog,
mathematical modeling, controlled drag
measurements in granular media, and bioinspired roboticsto address this question
(1). They find that properly coordinated tail
movements make locomotion efficient when
Bild Wissen Gestaltung: Ein Interdisziplinres Labor,
Exzellenzcluster der Humboldt-Universitt zu Berlin,
10099 Berlin, Germany, and Institut fr Biologie,
Humboldt-Universitt zu Berlin, 10115 Berlin, Germany.


limb motion is suboptimal and substrates are

challenging. Thus, the tail may have helped
stem tetrapods to move on land. The work
exemplifies a move in paleontology toward
increasingly interdisciplinary research (2).
In the Late Devonian and Early Carboniferous (about 360 million years ago),
an important yet only partly understood
evolutionary transition took place. Before
this time, stem tetrapods may have almost
never left the water. During the transition,
the first truly land-living forms appeared,
some of which eventually gave rise to all
modern tetrapods.
Salamanders with alternating limb
movements and hindlimb-generated propulsion are regarded as suitable modern
analogs for these earliest fully terrestrial

tetrapods. However, Pierce et al. recently

assessed limb joint mobility in the stem tetrapod Ichthyostegaa well-known species
from the Late Devonian with fully evolved
limbsand concluded that its hindlimbs in
particular were likely unsuited to effectively
generate propulsion on land (3). Hence, the
salamander model appears not to be adequate for the more aquatic Devonian stem
tetrapods in the rare cases when they ventured out of the water.
Pierce et al. proposed a crutching locomotion powered by simultaneous forelimb
movements for Late Devonian stem tetrapods (3). A similar mode of locomotion is
used by modern mudskippers, bony fishes
that spend a considerable amount of time on
land (see the photos). Further adding to our
understanding of mudskipper locomotion,
McInroe et al. now show how these fishes use
their tail to improve locomotor robustness.
However, locomotion in Ichthyostega may
have been different from that of other Devonian stem tetrapods. Furthermore, the
reconstruction of locomotor characteristics
from studying limb joint range of motion
is generally challenging (4), and the SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS




it needs to be shown that its predictions

closely match experimental data of the biomechanics measured in modern analogs.
Only if this is achieved should the model be
used to infer characteristics of a fossil (12).
Using bioinspired robotics as well as animations or even sophisticated musculoskeletal modeling (13) to elucidate function in
extinct organisms has further advantages.
Optimization criteria can be used to find
plausible parameter combinations without
user interference (10, 11). The results become less dependent on subjective opinion,
and future insights can more readily be incorporated into existing models. It is this
integrative methodology, exemplified by
McInroe et al.s study, that fosters increasingly interdisciplinary analyses of
function in fossils.
Whether or not mudskippers tail
use is an adequate modern analog
for stem tetrapods remains debatable. It can only be confirmed by
identifying morphological correlates
that are present both in modern
analogs and in fossils. McInroe et al.
do not resolve this issue, but rather
contribute to a more general understanding of the mechanism of coordinated tail use during crutching
locomotion on soft substrates. The
authors were interested in the overarching principle of tail use, and
their models serve as templates. In
contrast, attempts to reconstruct the
function of a structure of a specific
fossil should use its anatomical deCoordinated tail use. Mudskippers use their fins and tail to
tails as anchors for the model (3, 14).
navigate difficult terrain while out of the water. McInroe et al. use
McInroe et al.s study shows that
biomechanical analysis of mudskipper motion, mathematical
recruiting expertise from different
modeling, drag measurements in granular media, and bioinspired
fields facilitates integrated modelrobots to investigate whether early land vertebrates also relied on
ing approaches to problems of form
their tails to move on land.
and function in extinct organisms.
This enables new hypotheses to be
model as their starting point, the authors
generated and existing ones to be tested in
propose that stem tetrapod tails may have
a transparent and reproducible manner. j
improved the robustness of locomotion on
challenging substrates, in a similar fashion
1. B. McInroe et al., Science 353, 154 (2016).
as seen in mudskippers.
2. D. Jablonski, Science 284, 2114 (1999).
Both in robots and mathematical models,
3. S. Pierce et al., Nature 486, 523 (2012).
4. P. Arnold et al., J. Anat. 225, 31 (2014).
parameter combinations can systematically
5. G. Niedzwiedzki et al., Nature 463, 43 (2010).
be varied to access the sensitivity of the re6. S. M. Kawano, R. W. Blob, Integr. Comp. Biol. 53, 283 (2013).
sult to variation in individual parameters (2,
7. H. M. King et al., Proc. Natl. Acad. Sci. U.S.A. 108, 21146
10)something that is not possible when
8. J. S. Anderson et al., PLOS ONE 10, e0125446 (2015).
only living organisms are used as analogs.
9. J. A. Nyakatura et al., Evol. Biol. 41, 175 (2014).
Bioinspired robots are increasingly used
10. J. R. Hutchinson, S. M. Gatesy, Nature 440, 292 (2006).
11. A. J. Ijspeert, Science 346, 196 (2014).
as heuristic tools to investigate animal
12. P. S. L. Anderson et al., Biol. Lett. 8, 119 (2011).
adaptive behavior (11). Similarly, computer
13. S. L. Delp et al., IEEE Trans. Biomed. Eng. 54, 1940 (2007).
animation has great power for generating
14. R. J. Full, D. E. Koditschek, J. Exp. Biol. 202, 3325 (1999).
hypotheses on motion using digital modACKNOWLEDGMENTS
els (10). However, both approaches face
I thank E. Amson, J. Wlfer, and B. Kilbourne for their insightthe same problem: how to decide which of
ful critique of this manuscript. Bild Wissen Gestaltung: Ein
Interdisziplinres Labor is supported by DFG grant EXC 1027.
the many possible modes of locomotion is
the most realistic (10). To validate a model,


sieves for gas
Metal-organic framework
materials enable efficient
separation of similar gases
By Jerry Y. S. Lin

eparation and purification are critical industrial processes for separating

components of chemical mixtures, and
these processes account for about half
of industrial energy usage (1). Gas mixtures of compounds with very similar
physical properties are particularly difficult
to separate. On pages 137 and 141 of this issue, Cadiau et al. (2) and Cui et al. (3), respectively, show that microporous materials can
be designed to have high adsorption capacity
and selectivity for particular hydrocarbons,
enabling energy-efficient separation.
Traditional gas separation technologies
use distillation, requiring repeated evaporation and condensation of the mixture, or
absorption with a liquid medium that requires cooling and heating a large amount of
nonactive solvent to complete a separation
cycle. Both technologies are energy intensive.
Newer technologies use solid separation media and are generally more energy efficient.
For example, membrane-based separation
can require 90% less energy than distillation
to separate propylenepropane mixtures (1,
4). The cornerstone of these newer separation processes is a solid adsorbent or membrane, often made of a microporous material
with a pore size smaller than 0.5 nm and a
large internal pore surface area (>300 m2/g).
In many microporous solid media used
as adsorbents or membranes (5), there is a
trade-off between adsorption capacity (or
permeability) and selectivity for separating
challenging gas mixtures, making it difficult
for the adsorption or membrane process to
achieve high separation efficiency. To improve
adsorption capacity while also maintaining
selectivity, scientists have modified the internal surface properties of porous media to
enhance interaction of the adsorbent surface
with a specific component and thus increase
the adsorption capacity for that component.
For example, the crystalline microporous
zeolite LiX provides good adsorption capacity for nitrogen over oxygen due to Li cation
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

est potential tetrapod trackways suggest a

mode of locomotion more similar to that
of salamanders (5). If the locomotion seen
in mudskippers is a more adequate model,
it needs to be shown how alternating,
hindlimb-driven (salamander-like) locomotion evolved later on (6, 7). More intermediate fossils are needed (8), but alternative
modern analogs should also be considered
(9). Moreover, the mostly aquatic Late Devonian stem tetrapods with highly limited terrestrial capabilities probably encountered
challenging inclined sandy or muddy banks
when moving onto land. This is the issue
that McInroe et al. address in their interdisciplinary study. Drawing from diverse lines
of argumentation and using the crutching


Highly selective
In this issue, Cadiau et al. report a pillared square-gridstructured MOF with a controlled aperture opening
for molecular sieving of propylene over propane (right). Cui et al. have designed similar structured MOFs with
optimized surface functionalities and cage sizes for preferential adsorption of acetylene over ethylene (left).

Tilt results in smaller aperture




(SiF6 )2

surface functionalities, such as exposed metal

centers (8) or amines (9), in MOF-74 families
to enhance CO2 adsorption capacity and selectivity over N2. Alternatively, their molecular sieving characteristics can be exploited, as
has been shown for the MOF material ZIF-8
for propylenepropane separation (10).
These MOF materials have improved separation characteristics compared to classical
materials, but they still suffer from the adsorption capacity versus selectivity trade-off
for gas separations. For example, for acetyleneethylene separation, the MOF-74 family with a high density of open metal sites
shows high acetylene adsorption capacity but
low separation selectivity (8). For the same
Chemical Engineering, School for Engineering of Matter,
Transport and Energy, Arizona State University, Tempe, AZ
85287, USA. Email:




organic linker containing one or two pyridines or pyrazines, that form the upper and
bottom squares. The squares are pillared
apart via (SiF6)2 anions (called SIFSIX). The
pore aperture size on the squares can be
controlled by varying the length and/or orientation of the organic linker. The surface
chemistry is determined by the pillar ions,
metals, and organic linker molecules.
Cui et al. synthesized four (SiF6)2-pillared
MOF materials with different pore sizes by
changing the length of the organic linker or
creating an interpenetrated polymorph structure. They obtained high acetylene adsorption
capacity on MOFs of appropriate pore size
by optimizing hydrogen-bond interactions
between acetylene molecules and strongly
basic (SiF6)2, van der Waals interactions between acetylene molecules and the organic
linker, and the guest-guest molecule inter-

action. These interactions are much weaker

for ethylene because it is far less acidic than
acetylene. They identify MOFs with very high
acetylene adsorption capacity and acetylene
ethylene selectivity (2.1 mmol/g and ~44.8 at
0.025 bar, both higher than for previously
reported materials). Variation in the MOF
pore size does not influence the kinetics of
acetylene and ethylene adsorption. Rather, it
influences the host-guest and guest-guest interactions and hence the equilibrium adsorption capacity and selectivity.
Cadiau et al. study the same class of MOFs
but take a different approach, designing an
adsorbent for propylenepropane separation
based on the molecular exclusion principle.
They focus on the pillar-structured MOF with
the smallest cage, using an organic linker
containing only one pyrazine (see the figure).
Furthermore, they replace the (SiF6)2 pillars
in the cage with somewhat bulkier (NbOF5)2
pillars (called NbOFFIVE). This causes tilting of the pyrazine molecule on the linker,
effectively reducing the aperture opening
from 0.50 nm [with (SiF6)2 pillars] to 0.30
nm. With strong host-guest and guest-guest
interactions for propylene, the small aperture
permits transport of the smaller propylene
molecules into and out of the MOF cage, but
excludes the slightly bulkier propane. Thus,
this MOF shows high propylene adsorption
capacity and unprecedented (essentially infinite) propylenepropane selectivity.
The two studies present a novel direction
for the design and synthesis of MOF-based
adsorbents for challenging gas separations.
The authors report some stability data, but it
remains to be shown whether the adsorbents
are stable under industrially relevant conditions. The adsorption kinetics should also be
studied, especially the kinetic effects of MOF
framework flexibility at high adsorbate loading. Industrial application requires pelleting,
which might affect the adsorption characteristics. Future studies should also aim to
integrate these MOFs into membranes for
continuous gas separation processes. j


D. S. Sholl, R. P. Lively, Nature 532, 435 (2016).

A. Cadiau et al., Science 353, 137 (2016).
X. Cui et al., Science 353, 141 (2016).
A. Motelica et al., Ind. Eng. Chem. Res. 51 , 6977 (2012).
Y. S. Lin, M. C. Duke, Curr. Opin. Chem. Eng. 2, 209 (2013).
R. T. Yang, Adsorbents: Fundamentals and Applications
(Wiley, Hoboken, NJ, 2003).
M. Eddaoudi et al., Science 295, 469 (2002).
E. D. Bloch et al., Science 335, 1606 (2012).
T. M. McDonal et al., Nature 519, 303 (2015).
C. Zhang et al., J. Phys. Chem. Lett. 3, 2130 (2012).
M. C. Das et al., J. Am. Chem. Soc. 134, 8703 (2012).
P. Nugent et al., Nature 495, 80 (2013).


I am grateful to the NSF (CBET-1160084 and CBET-1511005) and

the Department of Energy (DE-FE0026435) for supporting work
on microporous materials and to Y. Liu and J. James for help in
preparing this Perspective.
10.1126/science.aag2267 SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016

separation, the ultramicroporous mixed

metalorganic framework (MMOF) family
has high selectivity but low acetylene adsorption capacity (11). The two studies in this issue overcome this trade-off problem through
crystal engineering that controls both surface
chemistry and pore size of the MOF materials. The resulting MOF materials have high
adsorption capacity and selectivity for separation of acetyleneethylene mixtures, as reported by Cui et al., and propylenepropane
mixtures, as shown by Cadiau et al.
The materials are based on a group of
MOF materials with a pillared square-grid
structure (see the figure) (12). The unit cell
consists of eight metal nodes, joined by an


nitrogen interactions (6). Selectivity can also

be achieved by varying the pore size of the
solid media to provide molecular sieving, as
in carbon molecular sieve (CMS) materials
(6). However, zeolites have a limited number
of crystal structures and CMS materials have
amorphous pore structures, making it hard
to generalize these approaches.
Metal-organic frameworks (MOFs) are microporous crystalline materials constructed
from transition-metal ions and bridging organic ligands (7). These materials can form a
vast variety of ordered structures, pore sizes,
porosity, and functional groups. MOFs can
be made into effective adsorbents or membranes for gas separation by introducing


Network analytics in the age of big data

By Nataa Prulj and Nol Malod-Dognin


e live in a complex world of interconnected entities. In all areas of

human endeavor, from biology to
medicine, economics, and climate
science, we are flooded with largescale data sets. These data sets
describe intricate real-world systems from
different and complementary viewpoints,
with entities being modeled as nodes and
their connections as edges, comprising large
networks. These networked data are a new
and rich source of domain-specific information, but that information is currently largely
hidden within the complicated wiring patterns. Deciphering these patterns is paramount, because computational analyses of
large networks are often intractable, so that
many questions we ask about the world cannot be answered exactly, even with unlimited
computer power and time (1). Hence, the
only hope is to answer these questions approximately (that is, heuristically) and prove
how far the approximate answer is from the
exact, unknown one, in the worst case. On
page 163 of this issue, Benson et al. (2) take an
important step in that direction by providing
a scalable heuristic framework for grouping
entities based on their wiring patterns and
using the discovered patterns for revealing
the higher-order organizational principles of
several real-world networked systems.
To mine the wiring patterns of networked
data and uncover the functional organization, it is not enough to consider only simple
descriptors, such as the number of
interactions that each entity (node)
has with other entities (called node
degree), because two networks can
be identical in such simple descriptors, but have a very different connectivity structure (see the figure).
Instead, Benson et al. use higherorder descriptors called graphlets
(e.g., a triangle) that are based on
small subnetworks obtained on a subset of
nodes in the data that contain all interactions
that appear in the data (3). They identify network regions rich in instances of a particular
graphlet type, with few of the instances of the
particular graphlet crossing the boundaries of
the regions. If the graphlet type is specified in

Network structures
The four networks shown have exactly the same size
(the same number of nodes and edges), and each
node in each network has the same degree (the
number of interactions with other nodes), but each
network has a very different structure.
Four triangles

Three squares

Two six-node cycles

Large, connected circle

advance, the method can uncover the nodes

interconnected by it, which enabled Benson et
al. to group together 20 neurons in the nematode worm neuronal network that are known
to control a particular type of movement. In
this way, the method unifies the local wiring
patterning with higher-order structural modularity imposed by it, uncovering higher-order
functional regions in networked data.
The importance of this result lies in its
applicability to a broad range of networked

RNAs and translated into proteins, which

adopt various three-dimensional structures
to carry out particular cellular functions.
Molecular interactions are captured by different high-throughput biotechnologies and
modeled with different types of networks.
Individual analyses of molecular networks
have revealed that molecules involved in
similar functions tend to group together in a
network and are similarly wired (13), leading
to better understanding of gene functions (6)
and molecular organization of the cell (7) and
to improved therapeutics (812).
However, each network type provides limited information about the phenomenon under study. For example, a disease is rarely the
consequence of a single mutated gene, or of a
single broken molecular interaction. Rather,
it is the product of multiple perturbations of
complex interactions within and across cells.
Network medicine couples network analytics
with data integration to mine the wealth of
complementary data and reveal common molecular mechanisms between seemingly unrelated diseases (811). By contrast, patients
with seemingly the same disease may have
very different molecular mechanisms of disease and reactions to treatment (e.g., cancer
heterogeneity) (811). Therefore, personalized
medicine aims at delivering individualized
therapies based on genetic and molecular
profiles of individual patients that may involve repurposing of known drugs to different patient groups, hence helping to ease the
pharmaceutical industry bottlenecks related
to the cost and time required to develop new
drugs (11, 12). Methods for network
data analytics and integration will
be fundamental to these nascent
areas, as full understanding can
only come from holistically mining
all available genetic, molecular, and
clinical data (11).
Holistic analyses of our interconnected world call for conceptual and methodological paradigm
shifts. Rather than analyzing a single data
source in isolation, such as aligning genetic
sequences (which has already revolutionized
our understanding of biology) (14), further insights will come from aligning all types of data
within a single frameworkthe data alignment. For example, all genetic and molecular
interaction data about a cell can be integrated
into the same computational framework, and
methods need to be developed for aligning

To mine the wiring patterns of networked

data and uncover the functional
organization, it is not enough to consider
only simple descriptors.

Department of Computer Science, University College London,

London, UK. Email:

data that we must understand to answer fundamental questions facing humanity today,
from climate change and impacts of genetically modified organisms, to the environment (4), to food security, human migrations,
economic and societal crises (3, 5), understanding diseases, aging, and personalizing
medical treatments (613). For example, the
cell is a complex system of interacting molecules, in which genes are transcribed into


8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

How can we holistically mine big data?


Holistic analyses of our

interconnected world
call for conceptual and
methodological paradigm
network types encoding the relationships
between climatic elements across geographical regions (e.g., wind speed, atmospheric
pressure, and temperature) (4), and holistic,
data-aligned analyses may help to explain this
complex, dynamic system and better predict
the effects of human-caused alterations. Mathematical formalisms capable of capturing the
intricacies of higher-order organization of the
data, along with the algorithms to compute
and extract information from those formalisms, should be developed and applied (15).
Extending the framework of Benson et al. to
finding higher-order structures within these
integrated and aligned data systems may be
a way forward. Computational issues remain
to be addressed, arising from large sizes, complexity, heterogeneity, noisiness, and different
time and space scales of the data. j

1. M. R. Garey, D. S. Johnson, Computers and Intractability:

A Guide to the Theory of NP-Completeness (Freeman,
New York, 1979).
2. A. R. Benson et al., Science 353, 163 (2016).
3. O.N. Yaverolu et al., Sci. Rep. 4, 4547 (2014).
4. K. Steinhaeuser, A. A. Tsonis, Clim. Dyn. 42, 1665 (2014).
5. P. Glasserman, H. P. Young, J. Bank. Financ. 50, 383 (2015).
6. R. Sharan et al., Mol. Syst. Biol. 3, 1 (2007).
7. K. Mitra et al., Nat. Rev. Genet. 14, 719 (2013).
8. A.L. Barabsi et al., Nat. Rev. Genet. 12, 1 (2011).
9. J. Menche et al., Science 347, 6224 (2015).
10. M. itnik et al., Sci. Rep. 3, 3202 (2013).
11. V. Gligorijevic et al., Proteomics 16, 741 (2016).
12. S. M. Strittmatter, Nat. Med. 20, 590 (2014).
13. D. Davis et al., Bioinformatics 31, 1632 (2015).
14. J. Alfoldi, K. Lindblad-Toh, Genome Res. 23, 1063 (2013).
15. S. Boccaletti et al., Phys. Rep. 544, 1 (2014).

This work was supported by the European Research

Council Starting Independent Researcher Grant (278212), the
National Science Foundation Cyber-Enabled Discovery and
Innovation Program (OIA-1028394), the Slovenian Research
Agency (ARRS) Project (J1-5454), and the Serbian Ministry of
Education and Science Project (III44006).



the Internet of Things
Photovoltaics can help to power remote sensors
and controllers at the edge of the Internet
By Richard Haight, Wilfried Haensch,
Daniel Friedman

he Internet connects billions of computational platforms of various sizes,

from supercomputers to smart phones.
However, the same types of data transmission can connect computational
resources to much simpler sensors
at the edge of the net that collect, analyze,
and transmit data, as well as controllers that
receive instructions. Devices deployed in the
environment, homes and offices, and even
our bodies would expand the number of connected devices to the trillions. This Internet
of Things (IoT) underlies the vision of smart
homes and buildings that could sense and
transmit their status and respond appropriately (1), or track and report on the state of
objects (vehicles, goods, or even animals) in
the environment. However, the practical implementation of the IoT has been relatively
slow, in part because all of these edge devices
must draw electrical power from their local
environment. We analyze the use of photovoltaics (PV) to power devices and help bring
the IoT to fruition.
Wide-scale deployment of devices to
remote or inaccessible areas while providing operational power in the absence of
wires would require harvesting of available energy to ensure long-term operation.
Indeed, one barrier to making an existing
building smart is the cost of installing
wired devices, so devices simply mounted
to walls and powered by ambient indoor
lighting are attractive for lowering installation costs. Whatever the specific form
of energy used (alternatives to PV include
kinetic, thermal, or wind energy), it must
have sufficient energy density and reliability. The power requirements are dictated
by the three components that process,
sense, and communicate data. An excellent
example of a system at the far edge of the
net is the ultralow-power chip developed
at the University of Michigan that incorporates much of the required functionality
and energy-harvesting infrastructure (2).
IBM T. J. Watson Research Center, Yorktown Heights, NY
10598, USA. Email:

An ultralow-power processor that might,

for example, be used in a smart phone (3)
would consume ~85 W of power when
operating at a clock frequency of 1 MHz.
Active power consumption in a processor is
given by CV 2F, where C is the capacitance, V is
the operating voltage of the processor,
and F is the clock frequency. Reducing the
operating voltage has the biggest benefit,
and energy-efficient processors have been
demonstrated to run with a supply voltage
of <400 mV at 0.83 MHz (4). The time constant of the systems monitored is often long
(milliseconds to minutes), so the required
clock frequency can be well below 1 MHz.
Intermittent operationwhere the processor can be woken to take a measurement,
store data, and later transmit it to a receiving stationwould save appreciable power.
Hence, ultralow-power operation of an edge
device is both required and achievable.
Sensors are another area where low power
is critical. As an example, consider an electrochemical sensor (5) used to monitor concentrations of toxic gases or other air-dispersed
pollutants. A typical sensor might consist of
a miniature electrochemical cell that draws
100 to 500 A of current at 1 to 2 V. The
total power consumed depends on whether it
operates continuously (as for a carbon monoxide monitor) or periodically (an outdoor
ozone monitor for air quality). Typical power
requirements for most sensors can range
from 100 W to >1 mW, depending on what
is being sensed.
Another critical component of an autonomously deployed edge device is its ability to
transmit data to other proximal devices or a
central location (hub) where the information
can be analyzed. There are several communication modalities that could be used that
operate near 1 V but differ in cost and distance range. A simple method would be to
transmit the data optically by modulating an
infrared lightemitting diode that operates
at milliwatt power levels (but limited distance range). Vertical-cavity surface-emitting
lasers are highly directional, providing longer range, but operate at several milliwatts.
Yet another communication mode involves
radio-frequency (RF) transmission being
developed for ultralow-power operation in SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016

these integrated cells within a new paradigm

of the cell alignment. Similarly, the worlds
economic system includes networks of trade,
financial exchanges, and investments, which
thus far have been studied individually (3, 5).
But a complete understanding of the origins
of wealth, crises, and economic recoveries
can only come from aligning and collectively
analyzing all of these layers of networked
economic and geopolitical data. Likewise, climatic measurements are captured by various

stored charge is depleted. The use of batteries as a voltage regulator avoids the need for
energy-consuming integrated-circuit voltage regulation. For a small computer-sensor
device, we may consider a 1-cm2 footprint
for a battery that is 0.1 cm thick; the maximum energy stored at an energy density of
200 Wh/liter for a lithium-ion battery (7)
is ~20 mWh. Given our estimates of power
consumption for processing and data storage
(~100 W), sensing (~1 mW), data transmis-

Powering edge devices

The Internet of Things can connect sensors deployed remotely, but these sensors must be powered. Photovoltaics
could power sensors both inside buildings and in the environment, as shown in this example of a smart ranch.

Suns rays

(Tracking and

Household functions
(Indoor light levels)

Crop status
Livestock location
and health status


Orchard hydration


Water levels

reduced, the viable separation distance to its

hub is limited.
Smart edge systems can also process data
locally and minimize the amount of communication with the central system. The
architecture of the network used to enable
hub-edge device communication will dictate data rates and distance specifications.
At present, ultralow-power devices (2) can
transmit a few meters, requiring a series of
other proximally located devices to relay data
over longer distances. Data transmission over
hundreds of meters or more is possible but
incurs a trade-off between the power cost of
processing and that for communication.
Given these power budget estimates, we
focus on energy harvesting, voltage regulation, and energy storage. One attractive
approach would be to merge thin-film PV
devices with an energy storage medium such
as a battery or supercapacitor. Batteries have
a distinct advantage in that they deliver current at a constant voltage, determined by the
particular electrochemistry used, until the

sion (~5 mW), and intermittent operation

totaling 1 hour per day, such an ultrasmall
battery would power an edge device for perhaps 3 days. In reality, systems deployed in
the field would need to last a year or more.
The foregoing simple analysis reveals the
acute need for energy harvesting to dramatically extend the life of the device in the
field. Light, as a nearly ubiquitous source of
energy, is an optimal choice. For PV cells,
the light absorber needs to be cheap, efficient, and, perhaps most important for field
applications, environmentally compatible,
because once deployed, it likely will not be
retrieved at a later date. The multielemental
kesterite Cu2ZnSn(S1xSex)4 (CZTS,Se) is the
most efficient Earth-abundant thin-film absorber available today. Although its power
conversion efficiency of 12.6% (8) is only a
little more than half that of CdTe and CdInGaSe (CIGSe), it does not contain metals
that are toxic (Cd) or expensive and relatively rare (In and Ga). The entire thin-film
PV device stack is only 2 to 3 m thick, just


1% that of the required 300 m of Si necessary to fully absorb ambient light.

There are also other important considerations in choosing a thin-film PV device to be
merged with a battery. The voltage produced
by the PV device at its maximum power point
(Vmpp) must exceed that necessary to charge
the battery. Although the standard voltages,
Voc (open-circuit voltage) and Vmpp, of a PV device are typically determined under full-Sun
illumination, called 1 sun (AM1.5) of ~100
mW cm2, charging of the battery must occur
under a wide range of illumination conditions and is a challenging aspect of PV-based
energy harvesting. Typical fluorescent lighting produces about 103 suns. At such low
light levels, the shunt resistance (9) of the cell
must be at least 500 to 1000 ohmcm2 to generate sufficient voltage to charge the battery.
Optimizing the band gap of the PV
absorber with the light source is another key
consideration. Indoor lighting from fluorescent or LED bulbs has a higher proportion
of blue light than the Sun. The band gap of
CZTS,Se can be correspondingly increased
by increasing its sulfur content. Emerging alloys of CZTS,Se, including (AgxCu1x)
ZTS,Se (10) with a variable band gap and
AgZTSe (11) with a 1.3-eV band gap, hold
promise for even higher efficiency than
traditional CZTS,Se because of suppression
of bulk defects. The band gap tunability
of CZTS,Se and CIGSe distinguishes them
from Si, GaAs, and CdTe. Higher voltages
can be achieved by connecting cells in series
or fabricating a tandem cell in which a top
PV cell with a larger band gap is deposited
on a bottom cell of a smaller band gap (12).
Truly self-powered, intelligent computers and sensors provide the ability to
communicate as a network and to gather
information on critically important aspects of even the most remote regions of
our planet. They also can help us to better
control our nearby environment to make it
more efficient and sustainable. j

1. N. Gershenfeld et al., Science 327, 1086 (2010).

2. P. Pannuto et al., in Proceedings of the 42nd Annual
International Symposium on Computer Architecture
(2015), pp. 629641.
3. ARM Launches Its Smallest, Lowest Power, Most Energy
Efficient Processor (press release);
4. B. Zhai et al., IEEE Trans. VLSI 17, 1127 (2009).
5. U. Guth et al., Meas. Sci. Technol. 20, 042002 (2009).
6. IMEC, IEEE 802.11AH Wi-Fi HaLow Radio in TSMC 40NM
7. J. B. Goodenough, K. S. Park, J. Am. Chem. Soc. 135, 1167
8. W. Wang et al., Adv. Energy Mater. 4, 130165 (2014).
9. T. J. McMahon et al., in Proceedings of the 25th IEEE
Photovoltaic Specialists Conference (1996), pp.
10. T. Gershon et al., Adv. Energy Mater. 6, 1502468 (2016).
11. E. Chagarov et al., J. Chem. Phys. 144, 104704 (2016).
12. J. Kim et al., Science 317, 222 (2007).
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

receive and transmit modes (6) with a total

power consumption of 6 to 8 mW and data
send and receive rates of 150 kilobytes per
second (kbps) to 4 Mbps. Devices are larger
for RF communication (typically centimeters
or larger) because of the antenna size. To
reduce edge device power needs, a hub system
with a central high-power transmitter would
only require low-power backscatter modulation from the edge device. Although the transmit power of the edge device is dramatically


Cite as: J. D. Boeke et al., Science

10.1126/science.aaf6850 (2016).

The Genome ProjectWrite

Jef D. Boeke,* George Church, * Andrew Hessel, * Nancy J. Kelley,* Adam Arkin, Yizhi Cai, Rob Carlson,
Aravinda Chakravarti, Virginia W. Cornish, Liam Holt, Farren J. Isaacs, Todd Kuiken, Marc Lajoie, Tracy
Lessor, Jeantine Lunshof, Matthew T. Maurano, Leslie A. Mitchell, Jasper Rine, Susan Rosser, Neville E.
Sanjana, Pamela A. Silver, David Valle, Harris Wang, Jeffrey C. Way, Luhan Yang
*These authors contributed equally to this work.
Corresponding author. Email:
The list of author affiliations is available in the supplementary materials.

The Human Genome Project (HGP-read) nominally completed in 2004 aimed to sequence the human genome and
improve technology, cost, and quality of DNA sequencing (1,
2). It was biologys first genome-scale project, and at the
time was considered controversial by some. Now it is recognized as one of the great feats of exploration, one that has
revolutionized science and medicine.
Although sequencing, analyzing, and editing DNA continue to advance at breakneck pace, the capability to construct DNA sequences in cells is mostly limited to a small
number of short segments, restricting the ability to manipulate and understand biological systems. Further understanding of genetic blueprints could come from construction of
large, gigabase (Gb)sized animal and plant genomes, including the human genome, which would in turn drive development of tools and methods to facilitate large-scale
synthesis and editing of genomes. To this end, we propose
the Human Genome ProjectWrite (HGP-write).
Responsible innovation
Genome synthesis is a logical extension of the genetic engineering tools that have been used safely within the biotech
industry for ~40 years and have provided important societal
benefits. However, recent technological advancementse.g.,
standardized gene parts, whole-genome synthesis, and clustered regularly interspaced short palindromic repeats
(CRISPR)Cas9 genome editing technology (3, 4)are revolutionizing the field (5). Some applications are controversial;
human germline editing in particular has raised intense
moral debate (6). As human genome-scale synthesis appears
increasingly feasible, a coordinated scientific effort to understand, discuss, and apply large-genome engineering
technologies is timely. HGP-write will require public involvement and consideration of ethical, legal, and social
implications (ELSI) from the start. Responsible innovation
requires more than ELSI, though, and involves identifying
common goals important to scientists and the wider public
through timely and detailed consultation among diverse

First release: 2 June 2016

We will enable broad public discourse on HGP-write;
having such conversations well in advance of project implementation will guide emerging capabilities in science
and contribute to societal decision-making. Through open
and ongoing dialogue, common goals can be identified. Informed consent must take local and regional values into
account and enable true decision-making on particularly
sensitive use of cells and DNA from certain sources. Finally,
the highest biosafety standards should guide project work,
and safety for lab workers, research participants, and ecosystems should pervade the design process. A priority will
be cost reduction of both genome engineering and testing
tools to aid in equitable distribution of benefitse.g., enabling research on crop plants and infectious agents and vectors in developing nations.
To ensure responsible innovation and ongoing consideration of ELSI, a percentage of all research funds could be
dedicated to these issues, enabling inclusive decisionmaking on the topics mentioned above (7). In addition,
there should be equitable distribution of any early and future benefits in view of diverse and pressing needs in different global regions. The broad scope and novelty of the
project calls for consideration of appropriate regulation
alongside development of the science and societal debates.
National and international laws and regulations differ and,
as in stem cell research, a major burden of responsibility of
setting standards rests with the scientists and their community. Existing stem cell research guidelines (8) may serve as
a useful template.
From observation to action
The primary goal of HGP-write is to reduce the costs of engineering and testing large (0.1 to 100 billion base pairs)
genomes in cell lines by over 1000-fold within 10 years. This
will include whole-genome engineering of human cell lines
and other organisms of agricultural and public health significance, or those needed to interpret human biological func-

(Page numbers not final at time of first release) 1

Downloaded from on June 2, 2016

We need technology and an ethical framework for genome-scale engineering

Pilot projects
Similar to other Gb-scale genomic projects, including HGPread, ENCODE (which aims to map genome functional elements), and Sc2.0 (which is synthesizing a heavily edited
yeast genome) (14, 15), HGP-write would be conducted in
phases with explicit goals and metrics. Each of the earlier
large-scale projects began with pilot projects focused on a
fraction of the genome, typically ~1%. For HGP-write, the
pilots should provide resources of immediate value for advanced biomedical research and/or biotech development.
Technology development will likely also occur early in the
project to propel large-scale genome design and engineering.
A series of pilot projects making use of very long DNA
sequences that are nonetheless short of a full genome are
anticipated: (i) synthesizing full gene loci with accompanying noncoding DNA to help explain still-enigmatic roles
of noncoding DNA variants in regulating gene expression,

First release: 2 June 2016

and leading to more comprehensive models for the role of

noncoding genetic variation in common human diseases
and traits; (ii) constructing specific chromosomese.g.,
chromosome 21or complex cancer genotypes to more
comprehensively model human disease; (iii) producing specialized chromosomes encoding one or several pathways
e.g., all genes needed to make a prototrophic human cell, or
pathways to transform the pig genome to make it more
amenable as source for human organ transplantation; (iv) a
potential transformation of gene therapy, with freedom to
deliver many genes and control circuits to improve safety
and efficacy, provided delivery challenges can be met. Indeed, many substantial and useful innovations may be realized in such stepping stone projects that are short of
whole-genome re-engineering but require substantial improvement in synthesis capacity of Mb- to Gb-sized DNA.
Both genome-wide and more modest changes could be tested for their impact on, e.g., organoid development and function in vitro, facilitated by ongoing progress in stem cell
differentiation and organ on chip technologies. Novel cell
culture technologies may, in some cases, be many times
more cost-effective and accurate than current wholeorganism testing.

Box 1. Some properties of ultrasafe cells with a pervasively re-engineered genome.

Virus resistancecan be conferred by systematically
recoding certain codons across all genes. Subsequent deletion of tRNA genes would generate a cell line resistant to
Improved cancer resistancetumor suppressor
genes could be made multicopy; genes like p53 could be
recoded to eliminate CpG dinucleotides that give rise to
hotspot mutations.
Other useful traitsdelete potentially deleterious
genes such as prion genes.
Improved genome stabilitycomprehensively eliminate endogenous repetitive selfish DNA elements.
Fail-safe securityprevent formation of germ cells,
e.g., by removing transcriptional regulators.
Applications go to cell line for stem cell therapies; robust production of biologics.
Additional pilot projects being considered include (v) using induced pluripotent stem cells (16) to construct an ultrasafe human cell line via comprehensive recoding of
protein-coding regions, and deletion of corresponding genome features to increase safety of such a cell line (see Box
1); and (vi) developing a homozygous reference genome

(Page numbers not final at time of first release)

Downloaded from on June 2, 2016

tionsi.e., gene regulation, genetic diseases, and evolutionary processes.

This goal is necessarily ambitious, since building a human genome at todays prices would cost more than HGPread (9) (see fig. S1). However, an expectation of HGP-write
will be to catalyze a sharp price drop as new technology development occurs apace with advancement of the project, as
with the cost of DNA sequencing in HGP-read. Small viral
(10) and bacterial (11) genomes synthesized from scratch
and organisms with recoded genomes (12) derived from
large-scale genome editing (13) have demonstrated the feasibility and utility of synthetic genomes. By focusing on
building the 3 Gb of human DNA, HGP-write would push
current conceptual and technical limits by orders of magnitude and deliver important scientific advances.
HGP-write will aim to address a number of human
health challenges. Potential applications include growing
transplantable human organs; engineering immunity to viruses in cell lines via genome-wide recoding (12); engineering cancer resistance into new therapeutic cell lines; and
accelerating high-productivity, cost-efficient vaccine and
pharmaceutical development using human cells and organoids. The project could encourage broad intellectual property access via patent pooling. Extreme cost-reduction is
feasible, as demonstrated by the $1000 genome grant program (2) as well as sharing of CRISPR tools from over 80
labs through Furthermore, because DNA synthesis, like sequencing and computation, is foundational
technology, HGP-write could also facilitate biological engineering of many organisms, accelerating R&D across a
broad spectrum of life sciences and supporting basic R&D of
new bio-based therapies, vaccines, materials, energy sources,
disease vector control, and nutrition.

Project launch and administration

The goal is to launch HGP-write in 2016 with $100 million
in committed support, from public, private, philanthropic,
industry, and academic sources from around the world. The
costs of the project lie not only in obtaining de novo synthesized DNA but in the assembly, integration, and functional
assays required to evaluate and understand the modified
genomes. Total project costs are difficult to estimate but
would likely be less than the $3 billion cost of HGP-read.
HGP-write could be implemented through one or more
centers [similar to Centers of Excellence in Genomic Science
(CEGS) and the Brain Research through Advancing Innovative Neurotechnologies (BRAIN) initiative centers] that will
coordinate and support formation and work of multiinstitutional and interdisciplinary research teams working
in a highly integrated fashion responsive to and engaged
with a broad public outreach.
We celebrate 2016the 25th anniversary of HGP-read
as a major step forward for human knowledge and health.
In this spirit, we look forward to the launch of HGP-write.

discussions. F.J.I. is a cofounder of enEvolv, Inc. G.C. has financial relationships

with Gen9, Editas, Enevolv, and Egenesis (companies directly related to this
article; for a full list of G.C.s financial relationships, see J.D.B. is a founder and on the Board of
Directors of Neochromosome Inc., owns stock in Sample 6, Inc., and is a
member of the Scientific Advisory Board of Recombinetics, Inc. A.H. has
investments in Autodesk Inc. G.C. is an inventor on patents and patent
applications filed by Harvard Medical School that cover synthesis, assembly and
testing of large DNAs..
Author affiliations
Fig. S1
Published online 2 June 2016

Downloaded from on June 2, 2016

bearing the most common pan-human allele (or allele ancestral to a given human population) at each position to
develop cells powered by baseline human genomes. Comparison to this baseline will aid in dissecting complex phenotypes such as disease susceptibility. The pervasive nature
of the required changes makes whole- or partial-genome
synthesis an efficient route to these goals.


1. International Human Genome Sequencing Consortium, Nature 431, 931 (2004).
Medline doi:10.1038/nature03001
2. C. W. Fuller et al., Nat. Biotechnol. 27, 1013 (2009). Medline doi:10.1038/nbt.1585
3. L. Cong et al., Science 339, 819 (2013). Medline doi:10.1126/science.1231143
4. P. Mali et al., Science 339, 823 (2013). Medline doi:10.1126/science.1232033
5. A. D. Haimovich, P. Muir, F. J. Isaacs, Nat. Rev. Genet. 16, 501 (2015). Medline
6. D. Baltimore et al., Science 348, 36 (2015). Medline doi:10.1126/science.aab1028
7. Presidential Commission for the Study of Bioethical Issues, New Directions: The
Ethics of Synthetic Biology and Emerging Technologies (Washington, DC, 2010).
8. J. Kimmelman et al., Nature 533, 311 (2016). Medline doi:10.1038/533311a
9. P. A. Carr, G. M. Church, Nat. Biotechnol. 27, 1151 (2009). Medline
10. K. J. Blight, A. A. Kolykhalov, C. M. Rice, Science 290, 1972 (2000). Medline
11. D. G. Gibson et al., Science 329, 52 (2010). Medline doi:10.1126/science.1190719
12. M. J. Lajoie et al., Science 342, 357 (2013). Medline doi:10.1126/science.1241459
13. F. J. Isaacs et al., Science 333, 348 (2011). Medline doi:10.1126/science.1205822
14. N. Annaluru et al., Science 344, 55 (2014). Medline doi:10.1126/science.1249252
15. J. S. Dymond et al., Nature 477, 471 (2011). Medline doi:10.1038/nature10403
16. K. Takahashi, S. Yamanaka, Cell 126, 663 (2006). Medline
This paper is the result of meetings held at NYU Langone Medical Center on 31
October 2015 and at Harvard Medical School on 10 May 2016. The authors
gratefully acknowledge the financial support of Autodesk, sponsor of the
meetings. We thank Debra Mathews (Berman Institute for Bioethics) for helpful

First release: 2 June 2016

(Page numbers not final at time of first release)


Supplementary Materials for

The Genome ProjectWrite
Jef D. Boeke,1* George Church,2* Andrew Hessel,3* Nancy J. Kelley,4* Adam Arkin,5 Yizhi
Cai,6 Rob Carlson,7 Aravinda Chakravarti,8 Virginia W. Cornish,9 Liam Holt,10 Farren J.
Isaacs,11 Todd Kuiken,12 Marc Lajoie,13 Tracy Lessor,14 Jeantine Lunshof,15 Matthew T.
Maurano,16 Leslie A. Mitchell,17 Jasper Rine,18 Susan Rosser,19 Neville E. Sanjana,20 Pamela A.
Silver,21 David Valle,22 Harris Wang,23 Jeffrey C. Way,24 Luhan Yang25
*These authors contributed equally to this work.
Corresponding author. Email:
Published 2 June 2016 on Science First Release
DOI: 10.1126/science.aaf6850
This PDF file includes:
Fig. S1
Author affiliations

Fig. 1. Synthesizing synthetic or semisynthetic genomes. A. Efficiency trends in

DNA sequencing (green) and synthesis of double-stranded DNA (dsDNA, blue) and
single-stranded DNA (ssDNA, red) over the past ~35 years. Double-stranded DNA, or
gene synthesis, has improved noticeably over the past ~10 years, but still lags behind
sequencing and ssDNA synthesis. The disruptive improvement in sequencing and
ssDNA (oligonucleotides) synthesis technologies has improved from multiplex and
miniaturization technologies in high-throughput DNA sequencing and oligo microarray
technologies, respectively. Commercial gene synthesis technologies relies on both oligo
synthesis (building blocks) and sequencing (validation of synthesis) technologies. B.
Graphical representation of four representative genomes benchmarked to the size of
the 3,000 MB human chromosomes: 9.5 kb hepatitis C virus (HCV) enlarged ~380,000fold, 1.1 MB Mycoplasma mycoides enlarged ~1,000-fold, 12 MB yeast enlarged 100fold.

As further support for the arguments in our paper, this is a (non-comprehensive)
sampling of precedents for projects that could take advantage of radical reduction in
cost of genome-scale synthesis and high-throughput cellular/organismal testing of
consequences. As with HGP-read, this effort need not be restricted to human but could
and should include mouse, pig, Drosophila melanogaster, Caenorhabditis elegans,
Arabidopsis thaliana, Saccharomyces cerevisiae, etc.
The bibliography, along with proposals for pilot projects, maybe found online at the
project web site
Author affiliations
Institute for Systems Genetics at New York University Langone Medical Center, New
York, NY, USA. Email:

Wyss Institute for Biologically Inspired Engineering at Harvard Medical School,

Massachusetts Institute of Technology, Boston/Cambridge, MA, USA. Email:

Autodesk Bio/Nano Research Group, San Rafael, CA, USA. Email:

Nancy J Kelley & Associates. New York, NY, USA. Email:

Department of Bioengineering, University of California, Environmental Genomics and

Systems Biology Division, E.O. Lawrence Berkeley National Laboratory, Berkeley, CA,
USA. Email:

Edinburgh Genome Foundry, School of Biological Sciences at University of Edinburgh,

Edinburgh, Scotland. Email:

Bioeconomy Capital, Seattle, WA, USA. Email:

McKusick-Nathans Institute of Genetic Medicine at Johns Hopkins University School of

Medicine, Batimore, MD, USA. Email:

Departments of Chemistry and Systems Biology at Columbia University, New York, NY,
USA. Email:

Institute for Systems Genetics at New York University Langone Medical Center, New
York, NY, USA. Email:

Department of Molecular, Cellular and Developmental Biology, Systems Biology

Institute at Yale University, New Haven, CT, USA. Email:


Science and Technology Innovation Program at Woodrow Wilson International Center

for Scholars, Washington, DC, USA. Email:

Department of Biochemistry at University of Washington, Seattle, WA, USA. Email:

Feinstein Kean Healthcare, Cambridge, MA, USA. Email:


Department of Genetics, Harvard Medical School, Boston, MA, USA; Department of

Genetics, University of Groningen, Groningen, The Netherlands. Email:


Institute for Systems Genetics, NYU Langone Medical Center, New York, NY, USA.

Institute for Systems Genetics, NYU Langone Medical Center, New York, NY, USA.
leslie. Email:

Department of Molecular and Cell Biology, University of California at Berkeley,

Berkeley, CA, USA. Email:

Edinburgh Genome Foundry, School of Biological Sciences at University of Edinburgh,

Edinburgh, Scotland. Email:

New York Genome Center; Department of Biology, New York University, New York,
NY, USA. Email:

Department of Systems Biology, Wyss Institute of Biologically Inspired Engineering at

Harvard Medical School, Boston, MA, USA. Email:

McKusick-Nathans Institute of Genetic Medicine at Johns Hopkins University School of

Medicine, Baltimore, MD, USA. Email:

Department of Systems Biology, Department of Pathology and Cell Biology at

Columbia University, College of Physicians and Surgeons, New York, NY, USA. Email:
24Wyss Institute for Biologically Inspired Engineering at Harvard Medical School,
Boston, MA, USA. Email:

eGenesis, Inc., Boston, MA, USA. Email:


Master manipulators
Emerging research on how parasites affect host behavior
fuels speculation about their influence on human history
By Shelley Adamo

the ability of the mosquito to take in a blood

meal, resulting in a perpetually hungry mosquito. It then bites multiple people, thereby
increasing the chance of parasitic transmission. There is even evidence that suggests
that Plasmodium can enhance the attractiveness of infected hosts to mosquitoes.
McAuliffe does not limit her discussion to
parasitology, referencing recent work on how
our gut microbes may influence our behavior

dult hairworms are aquatic, but their

larvae burrow into the abdomens of
terrestrial arthropods, like the cricket,
and freeload on nutrients derived
from unsuspecting hosts. This represents a potential problem, because
crickets are not aquatic animals. By mechanisms still unclear, once the hairworm is
ready to transition to the adult life
stage, it induces the cricket to jump
into water, where the worm is released
and the cricket drowns.
In her new book, This Is Your Brain
on Parasites, Kathleen McAuliffe examines the unusual and often dramatic
ways that parasites and microbial manipulators can influence the behavior of
their hosts, raising the question of how
much control we have over our own behavior. After reading the book, you may
come to the conclusion that it is actually far less than you once thought.
The book begins with vivid descriptions of parasites that hijack a range of
animal hosts. Like hairworms, many
of these master manipulators cannot
complete their life cycle unless their
Mice infected with toxoplasmosis no longer fear cats.
host can be compelled to behave atypically. For example, the larva of the parasitic
and exploring how the presence of parasites
wasp Hymenoepimecis argyraphaga, which
may have shaped the evolution of human
incubates in the abdomen of the orb-weavbehavior. Parasites exert a powerful selecing spider Plesiometa argyra, induces the
tive force, and behaviors that helped humans
spider to build a cocoon for it from the spiavoid infection should still be with us today,
ders own web before killing and consuming
she argues. Some evolutionary psychologists
its spider host.
suggest that human behaviors such as mate
The tales soon take on a more serious note,
choice, whether a society develops democas McAuliffe reveals how some parasitic maracy, and the decision to go to war may all be
nipulators are also implicated in human disinfluenced by pathogen pressure.
eases. For example, some pathogens, such as
One of the strengths of the book is McAuthe Plasmodium species that cause malaria,
liffes exploration of alternative explanations
enhance disease transmission by manipulatfor counterintuitive theories. This presentaing their insect vectors. They do this by reduction style alerts the reader that some of the
ing the amount of anticoagulants released by
ideas she recounts are imaginative speculaa mosquito during a blood meal. The insect
tion and not considered consensus. For readrelies on these anticoagulants to keep the
ers who prefer an authoritarian guide, the
blood flowing. In this way, plasmodia reduce
caveats and alternative perspectives may be
unsatisfying. However, such an approach is
preferable to the sensationalism that often
The reviewer is at the Department of Psychology and
surrounds work in this field.
Neuroscience, Dalhousie University, Halifax, NS B3H 4R2,
Canada. Email:
Sometimes McAuliffe simply presents op128

Houghton Mifin Harcourt,

2016. 292 pp.

posing views on a topic without comment,

leaving readers to decide how much weight
to give each argument. One example can be
found in her presentation of Toxoplasma
gondii, which causes rats and mice to become
attracted to cat urine instead of repulsed by
it. This strange attraction increases the odds
that an infected rodent will be eaten by a cat,
the parasites definitive host.
Humans harbor T. gondii, too: About 20%
of people in the Western world are infected.
Researchers have wondered whether a
parasite capable of altering neurotransmitter levels in the brain of a rodent
could also affect human brains and
behavior. Unfortunately, the only way
to know definitively whether a human
is infected with T. gondii parasites is
by autopsy. As a proxy, researchers often use the presence of specific markers in the blood to demonstrate that a
person has had contact with T. gondii.
However, research has shown that only
~50% of rats deliberately infected with
T. gondii end up with the parasites in
their brains. If a similar ratio exists in
humans, then there may be as much as
a 50% error rate in the assignation of
people into infected versus uninfected
groups. McAuliffe does not elaborate
on how this could represent a serious
problem for human work, but her decision to
include this information encourages readers
to make this inference on their own.
McAuliffe interviewed a wide array of
scientists working in the fields of ecoimmunology, psychoneuroimmunology, and
evolutionary psychology. Their stories demonstrate the importance of perseverance (and
serendipity) in science and the critical role
mentors can play in a career.
Scientists have much more to learn about
the neurological and behavioral effects that
parasites and other organisms can have
on hosts. Unfortunately, a cursory glance
through the web suggests that there are more
neurobiologists writing about fictional zombies than there are studying the secrets of
real-life zombie-makers. Perhaps this book
will inspire more researchers to pursue this
intriguing avenue of research.
10.1126/science.aaf9539 SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS

Downloaded from on July 8, 2016


This Is Your
Brain on Parasites
How Tiny Creatures
Manipulate Our Behavior and
Shape Society
Kathleen McAulife


B O OKS et al .


A far-reaching history explores the cultural, economic,
and political effects of earthquakes
By Sebastiano DAmico


espite advances in monitoring and

modern infrastructure, earthquakes
continue to represent one of the
most serious risks to health, safety,
and economic viability in many parts
of the world. This is, in part, because
science still cannot predict the exact time
and place where an earthquake will strike.
Yet, despite our insistence on treating such
catastrophes as acts of God, the truth is
that we humans seek out areas prone to major seismic activity. Indeed, more than half
of todays largest cities lie in areas that are
prone to earthquakes.
Traditionally, when scientists discuss
earthquakes, we talk about geology and
seismology, infrastructure and engineering, as well as management strategies for
minimizing human and material losses.
Nevertheless, the study of an earthquakes
effect on the social and cultural elements of
a community can help us understand how
different societies have evolved and adapted
over time and how cities have built up their
relative capacity to withstand future large
seismic events.
In fact, as Andrew Robinson describes
in his new book, Earth-Shattering Events,
the effects of an earthquake can reverberate throughout a societys identity. In some
cases, they have played a catalyzing role in
The reviewer is at the Department of Geosciences,
University of Malta, Tal-Qroqq, Msida, Malta.

Earth-Shattering Events
Earthquakes, Nations, and
Andrew Robinson
Thames and Hudson,
2016. 256 pp.

the evolution of urban and architectural

style and have irrevocably altered the communities in question.
From the first pages, Robinson captures
the readers attention with a whirlwind
summary of some of the more memorable
earthquakes in recorded history: the devastation of Lisbon in 1755; the 1960 Chilean
earthquakethe strongest in recorded historywhich caused a destructive tsunami
and unleashed more than 20,000 times the
energy of the atomic bomb dropped on Hiroshima; the earthquake in Tangshan, China,
that killed 750,000 in 1976.
The chronological organization of EarthShattering Events allows the reader to easily
follow the evolution of our understanding of
earthquakes. The book opens with descriptions of several memorable incidents that
occurred before the advent of seismology,
describing, for example, the panic that beset
London when an unprecedented five major
earthquakes shook Britain in 1750. The Lisbon earthquake in 1755 destroyed the city,
caused extensive damage in the nearby areas, and contributed to the birth of modern
Earthquakes have fostered changes in
political environments as well as served to


promote industrial and social growth. For

instance, Robinson suggests that the Tangshan earthquake, the most lethal earthquake
of the 20th century, may have struck the last
blow to the flailing Chinese Cultural Revolution. By contrast, he argues that the Gujarat
earthquake in 2001 further depressed the
economy of an already poor region, thereby
reinforcing authoritarianism in India. But
in both cases, the respective governments
used the earthquake to develop new infrastructures and catalyze industrial growth.
Similarly, after the 1906 destruction of San
Francisco, the city became an important
cultural center with a flourishing high-tech
industrial area that lies on the San Andreas
Faultone of the most active systems in
North America.
The book concludes with a description of
events that recently caught worldwide attention. In 2011, Japan experienced the largest
earthquake in its history. Initially estimated
at a magnitude of 7.9, the earthquakes true
magnitude9.0wasnt announced until
nearly 24 hours after it occurred. Upgraded
warnings about the resulting tsunami
heights failed to reach many coastal communities in time, and nearly 19,000 people
perished. It was the subsequent failure of
the Fukushima nuclear power plant, however, that garnered the most attention, with
several media outlets broadcasting uninterrupted coverage of the disaster.
Robinson clearly articulates the difference
between earthquake hazard and risk. The
first depends on the magnitude and location
of likely earthquakes, as well as how often
they occur, whereas the second quantifies
the potential damage an earthquake could
inflict on buildings or urban centers. The
latter is strongly linked to the concept of
vulnerability and the importance of seismic
retrofitting. Robinson places particular emphasis on the fact that the damage caused
by past earthquakes is very often neglected
when new structures are planned, resulting in modern cities with lax building codes
and few or no contingency plans, especially
in areas that are perceived as low-seismicity
regions. Quoting the engineering seismologist Nicholas Ambraseys, Robinson writes,
Earthquakes dont kill people; buildings do.
In conclusion, this book places earthquakes into proper historical perspective.
It will make for an enlightening read, both
for scholars and experts in seismology and
geosciences and for readers with an interest
in history.
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

A half-century after the largest recorded earthquake

struck Chile, a powerful 8.2-magnitude quake hit off
the countrys Pacific coast in 2014.


Edited by Jennifer Sills

Science stands by
2009 sheries study
THE EDITORS OF Science have become

Marcia McNutt

1. B. Worm et al., Science 325, 578 (2009).



Insufcient research
on land grabbing
OVER THE PAST decade, an unprecedented

boom in land transactionscommonly

referred to as land grabbinghas occurred
globally. At least 45 million hectares of
land have changed hands through concessions, long-term leases, and ownership
transfers (1, 2). Driven by volatility in
agricultural commodity prices, interest
in biofuel production, and eagerness of
governments to pursue economic development, transnational and domestic investors
have acquired land throughout the global

The company Saudi Star Agricultural Development transports water on the land it has bought in Ethiopia.

South (1). Resulting changes in control over

land threaten existing vegetation cover and
forests, especially where the new owners
successfully implement commercial agricultural production. Changes in control over
land can support greater agricultural output, but research on the subject has mostly
raised urgent concerns about transactions
leading to displacement of local livelihoods
and populations (3), and compromised
ecosystem services (4).
The efectiveness of research on land
transactions, however, is hobbled by three
problems. First, global data sets on land
transactions underestimate the total number of transactions. Ofcial statistics for
Ethiopia, Peru, and Cambodia show that the
numbers of transactions are consistently
underestimated in global data (5). Second,
it is difcult to calculate how much commercial agriculture is taking place because
not all of the transacted land is being
developed. Many investors, witnessing the
rise in large-scale land transactions, have
made speculative investments in land that
they hope later to sell for a prot. In many
other instances, local residents living in
and near transacted lands have resisted the
implementation of commercial agricultural
practices (6). Third, ndings reported in
the literature rest largely on samples that
are not statistically representative of the
variation in factors such as geography,
socioeconomic diferences, and contractual
arrangements that inuence outcomes.
Instead, they focus on social dimensions
of transaction outcomes such as fairness
and inequality (7). These are important to
consider, but the current dominant focus on
social outcomes needs to be supplemented
with a greater consideration of ecological
outcomes and efects on agricultural output.
A deeper understanding of land transaction outcomes requires studies that
are more representative of the range of
transactions. More systematic attention to
case selection and causal efects of tenure


changes is necessary to address research

limitations (8). Improved representation
will also enable more robust estimates of
social, economic, and ecological efects of
transactions (9).
Chuan Liao, Suhyun Jung, Daniel G.
Brown, and Arun Agrawal*
School of Natural Resources and Environment,
University of Michigan, Ann Arbor, MI 48109, USA.
*Corresponding author. Email:

1. B. White, S. M. Borras, R. Hall, I. Scoones, W. Wolford, J.

Peasant Stud. 39, 619 (2012).
2. B. Zagema, Oxfam Policy Pract. Agric. Food Land 11, 114
3. M. A. Wendimu, A. Henningsen, P. Gibbon, World Dev. 83, 84
4. K. F. Davis, K. Yu, M. C. Rulli, L. Pichdara, P. DOdorico, Nat.
Geosci. 8, 772 (2015).
5. Large-Scale Land Transactions as Deivers of Land-Cover
Change in Sub-Saharan Africa (
6. R. Hall et al. J. Peasant Stud. 42, 467 (2015).
7. K. Nolte, L. Voget Kleschin. World Dev. 64, 654 (2014).
8. S. M. Borras, C. Kay, S. Gmez, J. Wilkinson, Can. J. Dev. Stud.
Rev. Can. Dtudes Dv. 33, 402 (2012).
9. A. Agrawal, Proc. Natl. Acad. Sci. U.S.A. 111, 3909 (2014).

Mexico struggles to
keep foreign grants
IN HER NEWS In Depth story Mexico

struggles to woo expat genome jocks (29

April, p. 507), L. Wade discusses the bureaucratic hurdles that are slowing research in
Mexico. We would like to highlight a hurdle
not mentioned in the story: Researchers at
the Mexican National Institutes of Health
(INSHAE) encounter obstacles when trying
to claim U.S. National Institutes of Health
subcontractor grants.
Although not always successful, as Wade
points out, the Mexican government has
gone to great efort to make funds available
for local researchers. More worrisome are
8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


Downloaded from on July 8, 2016

aware of allegations of conict of interest

that Greenpeace has lodged against Ray
Hilborn for failing to disclose previous
industry funding in connection with his
co-authorship of the 2009 Science paper
Rebuilding global sheries by Worm et
al. (1). In fact, it is not possible to conclude
that Dr. Hilborn or other co-authors were
personally decient in their disclosures. At
that time, journal policy allowed a single
lead author to declare conicts of interest on behalf of all co-authors, a practice
that has since been discontinued. Current
journal practice requires each author individually to declare any and all relationships
(nancial or otherwise) that could constitute a real or perceived conict of interest.
Although the policies in place in 2009 may
not have resulted in the transparency we
have come to expect in 2016, the editors
of Science stand by the basic conclusions
reached in the Worm et al. paper. A group
of international sheries researchers came
together from diferent perspectives to reach
consensus on the status of the best-studied
marine ecosystems worldwide and agree on
solutions and challenges to rebuilding sheries stocks and achieving sustainable yields.


Christopher E. Ormsby, Santiago vilaRos, Gustavo Reyes-Tern*

Instituto Nacional de Enfermedades Respiratorias,
Mexico City, DF 14080, Mexico.
*Corresponding author.

atmospheric oxygen concentrations


1. Legal Notice [Ofcio] DGRI/685/2014 to the Head of the

INSHAE (2014);
pdf [in Spanish].


Comment on Long-term climate
forcing by atmospheric oxygen
Colin Goldblatt

Poulsen et al. (Reports, 12 June 2015, p.

1238) argued that lower atmospheric
oxygen levels during the Phanerozoic
would have given a warmer climate.
However, radiative and atmospheric
structure changes under lower pressure
both cause cooling, making their result
unusual in that a hierarchy of models gives
opposing results. Scrutiny of how radiative
and cloud processes were represented, and
a mechanistic explanation of the results,
are required.
Full text at

Response to Comment on
Long-term climate forcing by

Christopher J. Poulsen, Clay Tabor,

Joseph White

Goldblatt argues that a decrease in

pressure broadening of absorption lines in
an atmosphere with low oxygen leads to an
increase in outgoing longwave radiation
and atmospheric cooling. We demonstrate
that cloud and water vapor feedbacks in
a global climate model compensate for
these decreases and lead to atmospheric
Full text at

Erratum for the Review Somatic mutation in
cancer and normal cells by I. Martincorena
and P. J. Campbell, Science 351, aaf5401 (2016).
Published online 4 March 2016; 10.1126/science.
Erratum for the Report Ancient Ethiopian genome reveals extensive Eurasian admixture in
eastern Africa (previously titled Ancient Ethiopian genome reveals extensive Eurasian admixture throughout the African continent) by
M. Gallego Llorente et al., Science 351, aaf3945
(2016). Published online 19 February 2016; 10.1126/

New Version!

Over 100 new features &

improvements in Origin 2016!



Over 500,000 registered users worldwide in:

6,000+ Companies including
120+ Fortune Global 500
6,500+ Colleges & Universities
3,000+ Government Agencies & Research Labs

Published by AAAS

20+ years serving the scientic & engineering community

Downloaded from on July 8, 2016

the recent regulations (1) that essentially

disqualify INSHAE investigators from
receiving foreign grants. To claim international grants, researchers must draft
extremely complicated legal documents
and treaties, which are then rejected
by both the Mexican and the U.S. collaborators legal department. In addition
to hindering Mexican research, these
requirements place a huge burden on our
U.S. collaborators, who have to complete
onerous administrative tasks to support
scientic research in Mexico.
These funds are required to perform
urgent studies aimed at controlling infectious diseases caused for example by
the HIV, Chikungunya, and Zika viruses.
Scientists at the INSHAE have both
experience and geographical opportunity.
To allow much-needed funds to reach
Mexican scientists, the Mexican government should introduce realistic and
compliance-friendly regulations.



Comment on Long-term
climate forcing by atmospheric
oxygen concentrations
Colin Goldblatt
Poulsen et al. (Reports, 12 June 2015, p. 1238) argued that lower atmospheric oxygen levels
during the Phanerozoic would have given a warmer climate. However, radiative and atmospheric
structure changes under lower pressure both cause cooling, making their result unusual in
that a hierarchy of models gives opposing results. Scrutiny of how radiative and cloud processes
were represented, and a mechanistic explanation of the results, are required.

oulsen et al. (1) present results from the

Global Environmental and Ecological Simulation of Interactive Systems (GENESIS)
global climate model (GCM) that show a
warmer climate under lower atmospheric
pressure. This contrasts to established radiativeconvective model (RCM) results in which lower
pressure causes cooling (2). Changes to the clearsky radiation under less pressure have a net cooling effect (in their paper, Poulsen et al. described
only the warming effects). Likewise, atmospheric
structure changes weaken the greenhouse effect,
causing cooling. Yet their GCM model output
shows a warming via less cloud reflection. This
raises fundamental questions: Are the clear-sky
processes correctly represented in GENESIS? If
they are, what is the physical mechanism by
which a negative clear-sky forcing induces a positive cloud forcing strong enough to reverse the
sign of the overall change?
Less atmosphere leads to less Rayleigh (molecular) scattering. This can be treated with a simple scaling analysis, given that the amount of
scattering is proportional to the number of molecules in the column and that the scattering cross
sections for N2 and O2 are similar. (The scaling
with surface density used by Poulsen et al. was
erroneous.) To estimate the change in absorbed
sunlight, consider a single-layer, scattering only,
model atmosphere over a black surface. Let the
direct solar beam incident at the top of the atmosphere (TOA) be Idir(t) = I0. Using Beers law,
the direct beam at the surface is found Idir(s) =
I0et, where optical depth (t) is directly proportional to the number of molecules in the atmosphere. The Rayleigh scattering phase function
is symmetrical, so equal amounts of light are
scattered up and down and Iscat(s) = Iscat(t) =
I0(1et). t = 0.1 gives a good fit to Earths atmosphere [Iscat(t) = 16 W m2; observations (3)
and models (4) show that Rayleigh scattering is
15 W m2 today]. Decreasing the number of molSchool of Earth and Ocean Sciences, University of Victoria,
Victoria, British Columbia, Canada. Email:


8 JULY 2016 VOL 353 ISSUE 6295

ecules by 11% would correspond to decreasing t

by 11%, reducing the amount of reflected sunlight at the TOA by 1.7 W m2. This would tend
to warm the planet slightly.
In their scaling analysis, Poulsen et al. (1) conflate Mie scattering by clouds with Rayleigh scattering by molecules. Quite obviously, there will
be less Rayleigh scattering when there are fewer
oxygen molecules. However, there is no a priori
justification for the number of cloud droplets
scaling with the number of air molecules. The
decrease of reflected sunlight of 16 W m2 suggested by the Poulsen et al. (1) scaling cannot be
The other consequence of fewer molecules is
less pressure broadening of the absorption lines
of radiatively active species like CO2, so that these
absorb less radiation. The natural width of absorption lines is very narrow, but molecular collisions widen them so that more radiation overall
is absorbed. Lower palaeopressure would weaken
the greenhouse effect and tend to cool the planet.
Changes to pressure broadening dominate over
Rayleigh scattering, so the net clear-sky radiative
effect is negative forcing and cooling (2).
Mean atmospheric structure is well approximated by a moist adiabat. Given that saturation
vapor pressure depends only on temperature, the
amount of water vapor does not change with
pressure, but less dry air molecules make the saturation mixing ratio of water larger. Thus, lapse
rate is lower (less steep), there is less temperature difference between the surface and the atmosphere, and the greenhouse effect is further
weakened (2).
A straightforward way to evaluate the combination of the above is to calculate the radiative
forcing, with fixed surface temperature and CO2
inventory but varying O2 abundance (Fig. 1). The
radiative transfer here was performed at lineby-line resolution using the Spectral Mapping
Atmospheric Radiative Transfer (SMART) model, with absorption coefficients calculated from
HITRAN2012 (High-Resolution Transmission
Molecular Absorption Database) (5) specifically

for these atmospheric profiles, so these results are

of reference accuracy. For a decrease in O2 to 10%
of the atmosphere, the radiative forcing at the
tropopause is 3.0 W m2, consisting of thermal
and solar contributions of 4.4 and +1.4 W m2
(note the dominance of pressure broadening
over Rayleigh scattering). To first approximation,
surface temperature change is proportional to
radiative forcing. With a typical climate sensitivity of 0.8 K/W m2), this implies that reducing O2
to 10% would lead to a planetary cooling of 2.4 K.
Given that a warming is seen in the GCM, one
must first ask whether the driving radiative processes are represented correctly. Away from modern conditions, errors in GCM radiative transfer
codes are, unfortunately, rather common (6, 7).
They have to be heavily parametrized for speed to
run a single profile in a fraction of a second (compared to ~20 min for SMART). In palaeoclimate,
we take these codes well away from the modern
conditions for which they were tuned. Accuracy
cannot be presumed, and testing the code in the
conditions of interest is a due-diligence step (8).
Poulsen et al. (1) did not do this verification for
the legacy radiation code, National Center for Atmospheric Research Community Climate Model 3
(CCM3), used in GENESIS.
The standard method for testing GCM radiation codes is off-line comparison to a line-by-line
code, comparing radiative fluxes calculated over
fixed atmospheric profiles, so that the accuracy
of the radiation code may be separated from other model components [see Collins et al. (6) for a
thorough discussion of this methodology]. In correspondence with Poulsen et al., before this comment, I compared results of SMART (Fig. 1) and
CCM3 (kindly run by the authors), but the authors
did not want their results included here.
Next, one asks whether the model cloud response accurately represents the real world. Generally, the problem of modeling change to marine
stratus is fraught (9, 10). Circumstantial evidence
of a potential problem comes from an intercomparison project that addressed clouds for a different palaeoclimate topic: GENESIS gave cloud
radiative forcing 15 to 20 W m2 higher than any
other model, which was linked to how cloud radiative properties were represented (11). These
should be independent of climate state, so evidence of anomalous warming from clouds raises
substantial concerns.
Climate modeling has developed through a
hierarchy of models: For the canonical problem
of temperature response to CO2 doubling, all model classes from the original RCMs (12) to modern
GCMs give an answer of the same sign and similar
magnitude. Departure from this precedent is an
exceptional result, which requires exceptional evidence. GCMs have the great utility of resolving the
interaction of radiative transfer, dynamics, and
the hydrological cycle. However, the results are
only meaningful if all individual components are
accurate for the conditions in which the model is
used. For Poulsen et al.s response to this Comment, I would lay down two challenges. First, to
demonstrate that clear-sky and cloud radiative processes are verified as accurate for the conditions SCIENCE

Downloaded from on July 8, 2016



at hand. Second, in the context of other climate

forcings indicating cooling, to provide a direct,
physically based, mechanistic explanation for why
lower pressure should lead to less low cloud.
1. C. J. Poulsen, C. Tabor, J. D. White, Science 348, 12381241
2. C. Goldblatt et al., Nat. Geosci. 2, 891896 (2009).
3. Y. C. Zhang, W. B. Rossow, A. A. Lacis, V. Oinas,
M. I. Mishchenko, J. Geophys. Res. 109, D19105 (2004).
4. C. Goldblatt, K. J. Zahnle, Clim. Past 7, 203220 (2011).
5. L. S. Rothman et al., J. Quant. Spectrosc. Radiat. Transfer 130,
450 (2013).
6. W. D. Collins et al., J. Geophys. Res. 111, D14317 (2006).
7. C. Goldblatt, T. M. Lenton, A. J. Watson, Quart. J. R. Met. Soc.
135, 619633 (2009).
8. To reduce the need for ad hoc testing, a Palaeoclimate
Radiative Transfer Intercomparison Project has been established;
9. B. Stevens, G. Feingold, Nature 461, 607613 (2009).
10. S. Bony, J.-L. Dufresne, Geophys. Res. Lett. 32, L20806
11. D. S. Abbot et al., Geophys. Res. Lett. 39, L20711 (2012).
12. S. Manabe, R. D. Wetherald, J. Atmos. Sci. 24, 241259 (1967).

This work was supported by a Natural Sciences and Engineering

Research Council of Canada Discovery grant to C.G.
21 October 2015; accepted 26 May 2016

Fig. 1. Forcing from a change in number of moles of oxygen in the atmosphere relative to standard conditions (nstd). Colors are red for thermal, green for solar, and black for net. Forcings are
positive downward, so a positive forcing warms everything below that level. Atmospheric profiles are
constructed as follows. Troposphere (surface to pressure, p = 104 Pa): a moist adiabat is followed from
a 289 K surface to 240 K, and from p (240 K), temperature varies linearly with pressure to 206 K at p =
104 Pa, resulting in a profile that approximates a global annual mean (GAM) profile well when surface
pressure is 105 Pa. Stratosphere (above p = 104 Pa): GAM profile. The CO2 mixing ratio is adjusted to
conserve moles of CO2 between profiles. Water vapor varies linearly with pressure from the surface to
the tropopause [similar to Manabe and Wetherald (12)].


8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




Response to Comment on Long-term

climate forcing by atmospheric
oxygen concentrations
Christopher J. Poulsen,1* Clay Tabor,2 Joseph White3
Goldblatt argues that a decrease in pressure broadening of absorption lines in an
atmosphere with low oxygen leads to an increase in outgoing longwave radiation and
atmospheric cooling. We demonstrate that cloud and water vapor feedbacks in a global
climate model compensate for these decreases and lead to atmospheric warming.

oulsen et al. (1) report the results of a global climate model (GCM) used to investigate
the climate response to changing atmospheric oxygen (O2) levels and find that a
reduction in O2 leads to surface warming
and enhanced precipitation as a result of radiative changes and cloud, lapse rate, and water vapor
feedbacks. The results of Poulsen et al. (1) differ
from previous estimates using a one-dimensional
radiative-convective model (RCM) that only crudely represents climate feedbacks (2), raising questions from Goldblatt (3) about the representation
of clear-sky radiation processes in the Global
Environmental and Ecological Simulation of Interactive Systems (GENESIS) GCM and the mechanisms that lead to atmospheric warming in the
model. We address these two points below.
The GENESIS 3 radiation code is based on the
NCAR Community Climate Model (CCM3) radiation scheme (4). As noted in Goldblatt, a standard
method for testing radiation codes in GCMs is

to calculate radiation fluxes using a stand-alone

offline version of the radiation code with fixed
atmospheric profiles. We have completed this
analysis for clear-sky conditions using the same
atmospheric profiles of temperature and water
vapor for 10 and 21% O2 scenarios as Goldblatt to
facilitate comparisons between GENESIS 3 and
Spectral Mapping Atmospheric Radiative Transfer (SMART) models. Clear-sky top of the atmosphere (tropopause) longwave and shortwave
responses are approximately 2.5 and 0.3 W m2,
smaller but in line with the results from the
SMART model (Fig. 1). Differences in the magnitude of responses obviously result from differences in the radiation code but also stem from
choice in insolation, zenith angle, and interpolation of the atmospheric profiles to the radiation
model. In sum, as expected, the GENESIS 3 radiation code captures reductions in longwave absorption through Mie scattering and shortwave
absorption through Rayleigh scattering.

Fig. 1. Clear-sky and total-sky profiles

of net shortwave (SW) and longwave
(LW) radiation differences between
cases with 10 and 21% O2, estimated
using the GENESIS 3 radiation model,
which is based on the CCM3 radiation
scheme. Clear-sky profiles are consistent
with responses simulated by Goldblatt
using the SMART model and indicate a
decrease in the net TOA radiation as a
result of increased LW loss associated with
a reduction in pressure broadening of
absorption lines of radiatively active gases.
In contrast, total-sky profiles that include
clouds show an increase in the net TOA
radiation. The radiation profiles are estimated using the stand-alone GENESIS 3
radiation scheme using atmospheric profiles of water vapor and temperature
provided to us by Goldblatt. For total-sky
estimates, cloud profiles come from
climatologies of the 21 and 10% O2
GENESIS simulations. Both sets of profiles assume a TOA solar input of 338 W m2 and a zenith angle of 60.


8 JULY 2016 VOL 353 ISSUE 6295

It is well known that clouds have a large influence on Earths radiation budget and that
cloud feedbacks can determine the climate response to a forcing (5). As a crude illustration of
the influence of clouds due to a reduction in
O2, we include results for total-sky conditions
using global-average cloud profiles from climatologies of GENESIS 10 and 21% O2 simulations. The total (cloudy and clear sky) top-ofatmosphere (TOA) response to a reduction in
O2, 1.4 and 5.1 W m2 in longwave and shortwave, respectively, is much different and the
net (longwave + shortwave) response is opposite in sign (positive) than the clear-sky response
(Fig. 1), confirming the importance and warming
effect of cloud feedbacks. We note that the response described here differs, mainly in the surface total-sky longwave response, from the GCM
response because the temperature and water
vapor profiles are from Goldblatt and do not
account for GCM-simulated water vapor and
lapse rate feedbacks.
As illustrated by our radiation scheme calculations, feedbacks are responsible for the warming response in GENESIS. The initial response to
lower O2 is an increase in surface shortwave forcing, a warming of the surface and boundary layer,
and a reduction in atmospheric stability. Lowlevel warming has two direct effects: a decrease
in low-level, mainly stratus, clouds and an increase in convective activity. Low-level stratus
cloud fraction is determined in the model by a
relative humidity threshold and the presence
of either vertical ascent (for frontal clouds) or
low-level stratification (for marine stratocumulus
clouds) (6, 7). With greater surface shortwave radiative heating under low O2, the low-level relative humidity rises slightly (<2%), whereas both
vertical velocities in regions of mean ascent and
low-level inversions decrease [see, e.g., figs. S3
and S5 in (1)]. The net result is a decrease in lowcloud fraction, which acts as a positive feedback
on surface shortwave forcing. Between equilibrium runs of 10 and 21% O2, this positive cloud
feedback leads to a net surface shortwave forcing
difference of 4.8 W m2.
Enhanced surface shortwave forcing under low
O2 also drives both deeper and more frequent
convection. (The overall increase in convection
and reduction in mean vertical velocities are confirmed by corresponding increases and decreases
in simulated global-average convective and largescale precipitation.) Convective moistening of the
lower and mid-troposphere is seen in higher
relative humidity (by up to 10%) and increases
the downward longwave flux at the surface by
9.2 W m2, acting as an additional positive feedback on surface radiative forcing.
In addition to positive feedbacks, there are two
weaker negative feedbacks: (i) the moistening
of the atmosphere scatters incoming shortwave
radiation, reducing the surface net clear-sky

Department of Earth and Environmental Sciences, University

of Michigan, Ann Arbor, MI 48109, USA. 2National Center for
Atmospheric Research (NCAR), Boulder, CO 80305, USA.
Department of Biology, Baylor University, Waco, TX 76798, USA.
*Corresponding author. Email: SCIENCE

Downloaded from on July 8, 2016




is less correct than the other GCMs considered in

that study; the authors of the study do not make
such a claim (9). Incidentally, Poulsen et al. is not
the first to have found that clouds may be important in explaining past warm climates [e.g., (1013)].
Poulsen et al. considers the climate system
response to changes in atmospheric mass and,
in this way, is a step forward. The study identifies
a new forcing and provides mechanisms that potentially mitigate climate model mismatches with
past climates. Nonetheless, we caution that the
results, as in all modeling studies, have some
element of model dependence and must be
confirmed through additional climate modeling
efforts that incorporate feedbacks. Our hope is
that the study motivates a closer look at the effects
of changes in atmospheric mass as a potential
climate forcing throughout Earth history.

1. C. J. Poulsen, C. Tabor, J. D. White, Science 348, 12381241 (2015).


C. Goldblatt et al., Nat. Geosci. 2, 891896 (2009).

C. Goldblatt, Science 353, 132 (2016).
J. T. Kiehl et al., J. Clim. 11, 11311149 (1998).
V. Ramanathan et al., Science 243, 5763 (1989).
P. J. Rasch, J. E. Kristjansson, J. Clim. 11, 15871614
J. M. Slingo, Q. J. R. Meteorol. Soc. 113, 899927
W. D. Collins et al., J. Geophys. Res. 111, D14317 (2006).
D. S. Abbot et al., Geophys. Res. Lett. 39, L20711
L. C. Sloan, D. Pollard, Geophys. Res. Lett. 25, 35173520
D. S. Abbot, E. Tziperman, Q. J. R. Meteorol. Soc. 134, 165185
L. R. Kump, D. Pollard, Science 320, 195 (2008).
C. J. Poulsen, J. Zhou, J. Clim. 26, 70037022 (2013).


Preparation of this response was supported by NSF Sedimentary

Geology and Paleobiology Program grant F035175 and NSF Marine
Geology and Geophysics grant F033663.

Downloaded from on July 8, 2016

shortwave forcing by 0.9 W m2, and (ii) the

reduction in low-level clouds allows greater longwave escape from the surface by 5.6 W m2.
The explanations above address the two concerns raised by Goldblatt. The feedback mechanisms above were detailed in Poulsen et al. With
that said, we agree that the description of scattering in Poulsen et al. was not as clear as it could
have been and should have been referred to as
total scattering (Rayleigh and Mie). We do not
agree with a series of generalizations made by
Goldblatt that are not relevant to Poulsen et al.
Briefly, the reference to (8) is inappropriate. That
study examines the differences in radiation code
response to atmospheric CO2, which is not at
issue here. Likewise, characterizing the GENESIS
GCM cloud forcing in a study of the ice-free midCretaceous based on snowball earth simulations
with globally prescribed surface ice is irrelevant.
There is no way to assess that the GENESIS cloud
response in the snowball Earth comparison study

21 December 2015; accepted 26 May 2016


8 JULY 2016 VOL 353 ISSUE 6295




Asymmetric division of clonal muscle

stem cells coordinates muscle
regeneration in vivo
David B. Gurevich, Phong Dang Nguyen, Ashley L. Siegel, Ophelia V. Ehrlich,
Carmen Sonntag, Jennifer M. N. Phan, Silke Berger, Dhanushika Ratnayake,
Lucy Hersey, Joachim Berger, Heather Verkade, Thomas E. Hall, Peter D. Currie*
INTRODUCTION: Mammalian skeletal muscle

harbors tissue-specific stem cells that are triggered

to replace damaged fibers after injury. Genetic
ablation of satellite cells in the mouse results in
a failure to regenerate muscle, which indicates
that these cells are the major (and possibly only)
mediators for repair of skeletal muscle. Further
evidence for the central role of satellite cells
in muscle regeneration comes from transplantation experiments with genetically marked cells,
which demonstrate that satellite cells are highly
proliferative myogenic precursors capable of self
renewal and the resumption of quiescence, properties deemed important in a cell population
responsible for muscle repair. Considerable in
vitro evidence, derived from cultured fibers and
myoblasts, is suggestive of a role for asymmetric
division in generating both a self-renewing immortal stem cell and a differentiation-competent
progenitor cell that proliferates and ultimately
replaces damaged muscle. However, asymmetric
division of satellite cells has not been documented

in vivo. Furthermore, considerable doubt remains over how accurately in vitro studies can
model satellite cell behavior, given that the isolation and culture of individual muscle fibers
and cells stimulates satellite cell proliferation.
Finally, it is not clear whether the environment
an activated satellite cell encounters in a single
fiber explant, or in culture, mimics the molecular and biophysical architecture of a regenerating muscle injury in vivo. Consequently, what
role, if any, the wound environment itself plays
in regeneration and self-renewal is difficult to
address in these systems.
RATIONALE: Using the optical clarity and genetic tractability of the zebrafish system, we
developed tools to track and image the regeneration of living muscle tissue after injury. Marking
muscle stem and progenitor cells with transgenes
and using long-term imaging and lineage-tracing
modalities enabled us to visualize cell movements
and behaviors during regeneration in vivo.

RESULTS: In vivo cell tracking permitted highresolution imaging of the entire process of muscle regeneration, from injury to fiber replacement.
Using this approach, we were able to determine
the morphological, cellular, and genetic basis for
zebrafish muscle regeneration. Our analysis identified a stem cell niche in the zebrafish myotome
that is equivalent to the mammalian satellite cell
system, revealing that this evolutionarily ancient
stem cell is probably present throughout the vertebrate phylogeny. Complex interactions were observed between satellite cells and both injured and
uninjured fibers within the wound environment.
Among the most notable of these was the identification of filopodia-like projections, emanating
from uninjuredfibers,which
adhere to and lasso the activated satellite cell to guide
Read the full article
it to the wound edge. Furtherat http://dx.doi.
more, we documented the
in vivo occurrence of asym..................................................
metric satellite cell division,
a process that drives both self-renewal and regeneration via a clonally restricted progenitor pool.

CONCLUSION: Asymmetric divisions occur

during in vivo muscle regeneration to generate
clonally related progenitors required for muscle
repair. This finding resolves a long-term debate
surrounding the existence of this mechanism of
stem cell self-renewal and muscle repair in vivo.
Our results also reveal the highly dynamic nature
of the wound environment, where uninjured fibers
at the wound edge play a crucial role in directing
differentiating progenitors to regions of the wound
that are most in need of new fiber addition.

The list of author affiliations is available in the full article online.

*Corresponding author. Email:
Cite this article as D. B. Gurevich et al., Science 353, aad9969
(2016). DOI: 10.1126/science.aad9969

Mechanism of in vivo muscle repair. (A to C) Muscle regeneration is clonal. Regenerating fibers (outlined in white) express the same color after
fluorescent lineage tracing, indicating clonal derivation from a single stem cell. Sagittal, transverse, and coronal sections are shown in (A) to (C), respectively.
(D) Regeneration dynamics in vivo. Quiescent satellite cells, activated upon injury, undergo asymmetric division, which results in self-renewing or proliferating
cells. Proliferative cells undergo myogenesis to generate de novo immature fibers.

8 JULY 2016 VOL 353 ISSUE 6295 SCIENCE

Downloaded from on July 8, 2016




Asymmetric division of clonal muscle

stem cells coordinates muscle
regeneration in vivo
David B. Gurevich,1 Phong Dang Nguyen,1 Ashley L. Siegel,1 Ophelia V. Ehrlich,1
Carmen Sonntag,1 Jennifer M. N. Phan,1 Silke Berger,1 Dhanushika Ratnayake,1
Lucy Hersey,1 Joachim Berger,1 Heather Verkade,2 Thomas E. Hall,1 Peter D. Currie1,3*
Skeletal muscle is an example of a tissue that deploys a self-renewing stem cell, the
satellite cell, to effect regeneration. Recent in vitro studies have highlighted a role for
asymmetric divisions in renewing rare immortal stem cells and generating a clonal
population of differentiation-competent myoblasts. However, this model currently lacks in
vivo validation. We define a zebrafish muscle stem cell population analogous to the
mammalian satellite cell and image the entire process of muscle regeneration from injury
to fiber replacement in vivo. This analysis reveals complex interactions between satellite
cells and both injured and uninjured fibers and provides in vivo evidence for the
asymmetric division of satellite cells driving both self-renewal and regeneration via a
clonally restricted progenitor pool.

dult muscle stem cells or satellite cells play a

critical role in amniote skeletal muscle repair.
Normally quiescent and located between the
mature muscle fiber and the overlying basal
lamina, satellite cells are characterized by the
expression of a number of specific markers, including the paired homeobox genes Pax3 and Pax7,
the myogenic regulatory factor (MRF) gene Myf5,
and the hepatocyte growth factor membrane receptor cMet (1, 2). Genetic ablation of Pax7-expressing
satellite cells from adult muscle results in a failure
to regenerate, demonstrating that satellite cells are
the major and possibly only mediators of murine
skeletal muscle repair (3). The ability of satellite
cells to self-renew has been demonstrated in transplantation experiments using genetically marked
cells (4). Transplantation of a single intact myofiber with its complement of less than 10 satellite
cells has been shown to produce more than 100
new myofibers containing 25,000 nuclei, as well
as a contribution to the host pool of satellite cells
from the donor tissue (4). Such experiments demonstrate that satellite cells are proliferative myogenic cells capable of self-renewal and reassumption
of quiescence, all of which are important requirements for a stem cell population responsible for
muscle repair. However, exactly how self-renewal
is regulated remains an intense area of investi-

gation. The majority of studies have focused on a

possible role for asymmetric divisions of satellite
cells in generating a self-renewing immortal stem
cell and a progenitor cell required for the generation of a differentiation-competent population
that proliferates and ultimately replaces damaged
muscle (59). Although considerable in vitro evidence derived from cultured fibers and myoblasts
is suggestive of a role for asymmetric division
during satellite cell self-renewal, asymmetric division of satellite cells has not been documented to
date in any in vivo environment. Furthermore,
considerable doubt remains over how accurately
current in vitrobased paradigms can model satellite cell behavior. This is because the in vitro
culture of muscle fibers and the isolation of individual cells is inherently an activating process
for satellite cells. Finally, it is not clear whether
the environment an activated satellite cell encounters in a single fiber explant or in culture mimics
the molecular and biophysical architecture of a
regenerating muscle injury in vivo. Consequently,
what role, if any, the wound environment itself
plays in regeneration and self-renewal is difficult
to address. Recent studies using intravital imaging
within injured and uninjured mouse muscle have
begun to examine these questions in vivo, but technical constraints have prevented the examination
of asymmetric cell division and self-renewal (10).

De novo muscle repair mechanisms

in zebrafish

Australian Regenerative Medicine Institute, Level 1, 15

Innovation Walk, Monash University, Wellington Road,
Clayton, Victoria 3800, Australia. 2School of Biological
Sciences, Building 18, Monash University, Clayton, Victoria
3800, Australia. 3European Molecular Biology Laboratory
Australia Melbourne Node, Level 1, Building 75, Monash
University, Wellington Road, Clayton, Victoria 3800,
*Corresponding author. Email:


In this study, we subjected transgenic zebrafish to

distinct muscle injuries and followed regeneration by simple optical inspection of muscle tissue.
Of the injury models tested (1113), the one that
proved to be the most rapid, reproducible, and

amenable to gene expression analyses was the

needle-stick larval injury. In this model, an epaxial
myotome at the end of the yolk extension of larvae
at 4 days postfertilization (dpf) is injured with a
single puncture, using a 30-gauge needle held at an
angle of 75 relative to the body axis (Fig. 1, B to D,
and fig. S1, A to E). This form of injury resulted in
the loss of about half of all fibers in the epaxial
myotome at all medial lateral levels at the point
of injury. The repair process can be directly monitored in the living larvae via the use of transgenes that mark muscle fibers and by using the
light-refracting properties of intact muscle sarcomeres (termed birefringence) to quantitate muscle
integrity. Our analyses revealed that needle-stick
muscle injury in zebrafish larvae results in the
rapid activation of endogenous muscle repair,
a process that plateaus and concludes by 7 days
postinjury (dpi) (fig. S1, A, B, and G, n = 20 fish;
fig. S1, C to E, n = 7 fish; and movie S1). Fiber loss
occurs immediately after injury and continues for
several days. There is little evidence of wound
contraction contributing to repair because the
wound site remains stable during the regenerative
process until it is repaired by new fiber addition
(fig. S2, n = 8 fish; and movie S2).
To study the cell biological basis of this repair,
we performed needle-stick injuries in fish that
were transgenic for both Tg(actc1b:mcherry) and
Tg(myf5:GFP) (GFP, green fluorescent protein),
which mark differentiated muscle fibers in red
and myf5 expression in green (Fig. 1, A to D, n =
12 fish; figs. S2 and S3, A to D, n = 38 fish; and
movies S1 and S2). Myf5 in amniotes is expressed
in both quiescent and activated satellite cells and
transit-amplifying progenitors during muscle regeneration and, consequently, marks an extended
phase of the regenerative process (2, 14). We
theorized, therefore, that the Tg(actc1b:mcherry)
and Tg(myf5:GFP) transgene combination may
allow us to image stem cell activation to muscle
differentiation in continuous time-lapse analyses.
In uninjured larvae, myf5-positive cells are located
throughout the myotome, interstitial to differentiated muscles fibers, and first appear at 4 dpf
(Fig. 1A). Injury induces a series of dynamic cell
shape changes within a proportion of myf5-positive
cells; these changes include the extension of long
filopodial processes into the wound, reminiscent
of the behavior deployed by muscle progenitors
within injured adult mouse muscle (10). Consequently, migrating myf5-positive cells exhibit an
elongated shape, entering the wound site from
multiple levels and positions within the injury
(movie S3), following their membrane extensions
into the wound site (Fig. 1, B to D, and movie S1).
This activated subset of myf5-positive cells is chiefly
characterized by their polarized, migratory phenotype, with activated cells often elongating and migrating over nonextended myf5-GFP cells that are
initially closer to the wound (Fig. 1, B to D). By
2 dpi, myf5-positive cells within the wound site
have begun to elongate as myocytes, with evidence
by 3 dpi of fully differentiated muscle fibers replacing those that have been lost (fig. S1, C to E,
n = 7 fish; and fig. S3, A to D). myf5 expression
peaks at 3 dpi and declines steadily, returning to
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



preinjury levels by 7 dpi, when regeneration is

complete (fig. S3, A to D).
To examine the nature of the de novo myogenic
event induced by muscle injury, we investigated
the expression of endogenous genes involved in
muscle repair in amniotes. Specifically, we examined the expression of pax7, which marks both
quiescent and activated muscle stem cells and
is the most universally accepted marker of satellite cells in postnatal murine muscle (5). We also
investigated the expression of the MRF genes myf5,
myoD, and myogenin, which are sequentially activated during regeneration in mammals (2). For all
myogenic regulators, expression is specifically induced at the injury site, in a temporal fashion consistent with their roles in controlling a local de

novo myogenic event after injury (12) (fig. S3, E

to H, n = 10 fish per group).
Furthermore, regenerating fibers initially express
an embryonic slow myosin heavy-chain (MyHC)
isoform during differentiation. This observation
reinforces the notion that different processes, distinct to those that occur during muscle growth,
coordinate muscle regeneration. Regenerating fibers express the slow MyHC1 gene (15) and protein (16), and quantitation of this expression (fig.
S1, H to R, n = 6 fish) as well as analysis of the
expression of the slow MyHC1:GFP transgene [Tg
(smyhc1:GFP)] during injury (Fig. 1, E to G, n = 14
fish), revealed that this is a transient expression
exhibited specifically by newly regenerating fibers.
Embryonic slow MyHC1 expression occurs despite

the fact that the injury encompasses a region of

the myotome that is mainly composed of fast
MyHC-expressing fibers (Fig. 1, E to G, and fig. S1,
H to R). Transient embryonic slow MyHC expression, independent of the predominant fiber type of
the injured muscle, is also a feature of regenerating fibers in amniote injury models (1720). We
therefore concluded that muscle injury in zebrafish results in activation of a specific de novo
myogenic program.
Stereotypical phases of muscle repair
in vivo
Next we induced muscle injuries via laser ablation of muscle fibers in the Tg(actc1b:mcherry)
and Tg(myf5:GFP) double-transgenic line. Laser


8 JULY 2016 VOL 353 ISSUE 6295

Downloaded from on July 8, 2016

Fig. 1. In toto imaging of muscle regeneration in vivo.

(A to D) Uninjured muscle [Tg(actc1b:mCherry), red] contains
myf5-positive cells [Tg(myf5:GFP), green], a subset of which
are activated postinjury and exhibit a highly bipolar extended
shape [arrowheads in (B) to (D)].The remainder of the myf5positive cells fail to be activated and do not migrate to the
injury (arrows). Representative images taken from n = 12 independent fish. Scale bar, 50 mm. (E to G) Newly regenerated
muscle initially expresses slow MyHC, shown by Tg(smyhc1:
GFP) (green), despite the injury predominately affecting
deeper fast MyHCexpressing muscle [Tg(actc1b:BFP), blue].
Representative images taken from n = 14 independent
fish. Arrows indicate slow MyHC-expressing fibers in the
fast muscle domain. Scale bar, 50 mm. (H) Stereotypical
phases of muscle regeneration. Maximum intensity projection
through Tg(myf5:GFP) and Tg(actcb1:mCherry) doubletransgenic larvae, postlaser wound injury, time-lapsed
over 48 hours after injury. Arrows mark cell divisions
and arrowheads marks the filopodial extension from the
adjacent uninjured muscle. T = time in minutes. Representative images taken from n = 8 independent fish. Scale
bar, 50 mm. (I) Quantitation of the length of the phases
described in (H). (J to N) Inhibition of filopodia prevents
regeneration. (J) Quantitation of birefringence after needlestab injury in 3-dpf larvae soaked in a low-concentration
range of cytochalasin D, shown to specifically inhibit filopodial
extensions in vivo and in vitro. All concentrations inhibit regeneration by 4 dpi; the first time point that a recovery in
birefringence is detected in untreated siblings. Error bars
indicate mean + SEM. Statistics: t test, two-tailed numbers
(n) are indicated at each time point and concentration.
****P < 0.0001. (K to N) Individual larvae followed over the
entire period of regeneration were imaged for birefringence
each day. (K to L) Control untreated siblings recover rapidly
form needle-stick injury. (M to N) Larvae treated with
50 mM cytochalasin D fail to regenerate and reestablish
birefringence by 4 dpi. (K to N) High-magnification views of
the regions boxed in (K) to (N), respectively. SCIENCE

ablation results in single muscle fibers or small

groups of fibers being ablated in a more defined area
of injury when compared with the needle-stick
injury used above. Regeneration of this laser injury is more rapid and occurs over a 3 days (fig. S1F,
n = 20 fish). This allows the entire wound to be captured at resolutions that can distinguish individual
cells. We were able to capture the entirety of the
regenerative process, from the initial injury to the
differentiation of replacement muscle cells. This
analysis, coupled with those described above,
allowed the definition and quantification of
specific phases in the regenerative processes (Fig.
1, H and I, and movie S4, n = 8 fish).
The initial phase (migration and contact) is
characterized by migration of the myf5-positive
cell populations into the wound area. Migration
of myf5-positive cells is limited to within the injured
myotome itself and the neighboring myotomes
on either side. Migration to the wound site can
occur from any position within these myotomes.
However, we never see dorsal-ventral traverse of
the horizontal myosepta for discrete wounds
within either the epaxial or hypaxial myotomes
where the septum remains intact after injury
(movies S1 to S7). Once in the wound, cells associate as rounded cells with severed and dying fibers
(Fig. 1, H and I, and movie S4, green arrows). In the
second phase (proliferation and activation), myf5positive cells undergo division to generate muscle
progenitors (Fig. 1, H and I, and movie S4, cyan
arrows), similar to events and behaviors exhibited
by myogenic progenitors in early muscle repair in
mice (10). Quantitation of the total number of
proliferative cells confirmed that muscle regeneration was associated with a rapid proliferative
burst during the 2 days immediately after injury
(fig. S6, A, C, and I, n = 6 fish). EdU pulse labeling
after needle-stick injury confirmed the proliferative nature of the myf5-positive cells detectable
within the wound site, in line with our time-lapse
observations of laser ablation injuries (fig. S6, F
and I, n = 6 fish). In the third phase (bipolarextension and interaction), myf5-positive cells
extend to a bipolar shape, increase myf5:GFP
expression, and lose contact with the dying cells
around them. Bipolar myf5-positive cells in the
wound then actively migrate toward each other,
extend filopodia to make contact, and then recoil
and move rapidly apart (Fig. 1, H and I, and movie
S4, blue arrows). Next, resident uninjured differentiated muscle fibers extend filopodial-like membrane protrusions, which adhere and lasso around
the activated progenitor cells (Fig. 1, H and I,
movie S4, yellow arrows, and fig. S4). Filopodia
can be observed in every time-lapse analysis of
sufficient length that we performed for both
types of injury: needle stick and laser ablation
(n = 8 fish, Fig. 1H, fig. S4). Filopodia are invariably associated with the arrival of the myf5positive cells into the wound site and extend to
contact them. The size and nature of the filopodia
are dependent on the location and distance between the uninjured muscle at the wound and
the position of the myf5-positive cells relative to
the muscle cell at the wound edge (Fig. 1H, movie
S4, and fig. S4). The timing and directionality of

the extending filopodia suggest that signals from

the migrating and dividing stem and progenitor
cells may elicit and guide the filopodia. Consequent filopodial retraction appears to guide the
lassoed elongating myocytes to align next to the
differentiated muscle fiber from which the contacting filopodia extends (Fig. 1, H and I, and movie S4).
To define the functional role of filopodia in the
regenerative process, we inhibited their formation
after injury. Low-dose cytochalasin D selectively
suppresses filopodia formation by capping actin
filament barbed ends while sparing the formation of lamellipodia, another F-actindependent
membrane extension process (21, 22). Furthermore, cytochalasin D has recently been applied
in vivo to specifically inhibit filopodia in developing mouse and zebrafish embryos (23, 24).
Using a similar approach, we applied low-dose
cytochalasin D to animals injured by needle-stab
and assayed regeneration under these conditions.
Addition of low-dose cytochalasin D had no overall effect on larval morphology and muscle integrity in uninjured animals (Fig. 1, M to N, n = 59).
However, injured animals showed a marked defect in regeneration upon cytochalasin D treatment (n = 17, 5 nM; n = 17, 25 nM; n = 19, 50 nM),
which suggests that impeding filopodia formation
leads to inhibited regeneration (Fig. 1, J to N).
After alignment against resident uninjured
fibers, myocytes elongate and initiate expression
of Tg(actc1b:mcherry), signaling the differentiation
of the myf5-positive progenitor pool into muscle
fibers that heralds the end of the final phase
(resident fiber capture and differentiation) (Fig. 1,
H and I, movie S4, white arrows). Thus, our analyses
have determined that muscle regeneration is associated with a series of stereotypical morphogenetic
interactions between cells at the wound site
and the activated stem cells.
Muscle regeneration is impaired in
myogenin mutants
Next we examined whether muscle regeneration
deployed a genetic program distinct from that required during embryonic myogenesis. In mice, the
MRF genes Myf5 and MyoD are individually required to direct regeneration of skeletal muscle
(25, 26). Mutations in each of the zebrafish MRF
genes (myf5, myoD, mrf4, and myog) have been
previously generated (27, 28). Using birefringence
(fig. S5, B and F, n = 10 fish per condition), gene
expression, and muscle differentiation markers
(fig. S5, C to E, n = 6 fish per probe per condition)
to systematically monitor regeneration in homozygous mutants, we determined that only homozygous myogenin (myog fh265) mutants possessed a
muscle-repair deficit (fig. S5). Myogenin is required
for the differentiation of muscle cells during embryonic amniote myogenesis and is consequently a
prenatal lethal mutation in mice (29). By contrast,
zebrafish myog mutants possess no apparent
defect in embryonic myogenesis, as myod and
myog act redundantly to coordinate embryonic
fast muscle myogenesis (27, 28). Similar numbers
of stem cells are present at the wound site initially, with the deficit in repair being accompanied by a subsequent defect in proliferation at the

injury (fig. S6, D, E, and H, n = 6 fish). To determine more precisely the phase of regeneration
that was deficient in myog mutants, homozygous
myog fh265 larvae transgenic for both Tg(myf5:
GFP) and Tg(actc1b:mCherry) were subjected to
laser ablation muscle injury and continuous timelapse imaging, as described above (Fig. 2A and
movie S5, n = 7 fish). myf5-expressing cells in
myog f h265-mutant fish migrate effectively to the
injury site. However, these cells failed to fully
extend and mature to a bipolar shape and were
eliminated from the wound site (Fig. 2, A and
B, and movie S5), consistent with the marker
analyses described above. Specifically, myog fh265mutant fish possessed dying myf5-expressing
cells only within the injury site (Fig. 2, C to H,
n = 6 fish), which indicates that myog acts as a
survival checkpoint for myoblasts involved in
muscle repair. This analysis therefore defines a
genetically distinct set of myogenin-dependent
cells that are specifically required for muscle
regeneration in zebrafish.
Specific stem cell populations control
muscle repair
Although these initial studies reveal that a genetically distinct population of cells contribute to
the regenerative process, they do not formally
eliminate the possibility that these cells arise
through dedifferentiation at the wound site and
consequently contribute to muscle regeneration
via the de novo myogenesis process described
above. To determine whether injured skeletal
muscles can undergo dedifferentiation and contribute to repair, we injected the construct actc1b:
nlsGFP into the Tg(actc1b:mCherryCAAX) transgenic fish, which results in mosaic genetically
marked myonuclei (green) in a fish with all muscle fiber membranes marked (red) (fig. S7 A). We
performed laser ablation of muscle cells containing
these green nuclei and showed that the nuclei of
injured muscle fibers are cleared from the site of
the wound and subsequently do not play a role
in repair via dedifferentiation (fig. S7, A to D,
n = 6 fish). To determine whether surrounding
uninjured myofibers were capable of dedifferentiating and playing a role in injury repair, we
performed needle-stick injuries to fish doubletransgenic for Tg(actc1b:Cre) and Tg(ef1a1:LOXPeGFP-LOXP-mCherry) (eGFP, enhanced GFP),
marking all tissue in the fish as green, except for
mature myofibers that expressed actc1b (marked
as red). Examination of these injured fish reveals
that the regenerate consists of green cells early
in the repair process, indicating that these cells
are not derived from dedifferentiated mature myofibers (fig. S7, E to G, n = 6 fish). Together, these
results show that skeletal muscle regeneration in
zebrafish does not largely occur by dedifferentiation
of injured or uninjured muscle cells at the injury
site. Skeletal muscle regeneration is therefore fundamentally different from the regeneration of the
other striated muscle type, cardiac muscle, which
in zebrafish is controlled by the reentry of cardiomyocytes into the cell cycle in response to injury (30).
We next looked for a source of potential stem
cells within the myotome. Although myf5 expression
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


encompasses the broadest temporal phase of an

activated stem cell in amniote muscle repair, it
occurs extensively throughout all phases of myogenesis and is not a marker specific for the stem
cell compartment. We therefore sought to define
markers that were specific for muscle stem cells
deployed during regeneration. Our analysis revealed that pax7a- and pax3a-positive cells are
evident deep within the myotome, beginning at
4 dpf in a similar location to the myf5 cells described
above (Fig. 3, A to F, n = 6 fish). However, these
genes are also expressed by cells of the external

cell layer (ECL), a progenitor pool for muscle growth

located on the superficial surface of the myotome,
evident during embryonic and larval periods.
Therefore, the expression of these genes is not
specific to a muscle stem cell niche (11, 12, 31, 32)
(Fig. 3, A and D). cMet is a marker of quiescent
satellite cells in postnatal rodent muscle and is
thought to be specifically required for the activation of satellite cells in response to injury (2, 33).
In line with its proposed function, cmet expression initiates at 4 dpf, later than either pax3 or
pax7; is expressed in only a small subset of the

Fig. 2. Myogenin activity is specifically required for muscle regeneration as a survival checkpoint
for proliferating myoblasts.
(A) Maximum intensity projection
of a myogfh265 mutant fish, doubletransgenic for Tg(myf5:GFP) (green)
and Tg(actcb1:mCherry) (red), timelapsed over 35 hours after injury.
t, time in minutes. myf5-expressing
cells (arrows) exhibit defects in
differentiation. Specifically, cells enter the wound site and initiate bipolar extension but fail to progress
to differentiation and die within
the wound. Representative images
were taken from n = 7 independent fish. (B) Quantitation of the
length of the phases described in
Fig. 1 for WT injuries, as related to
the phases observed in the myogfh265
mutant context, from (A). (C to F)
Maximum intensity projection of
needle-stickinjured WT and myogfh265 mutant fish, transgenic for Tg(myf5:GFP) (green) and stained
with TUNEL (red). myogfh265 mutant fish possess myf5-positive cells that are also positive for TUNEL
staining (arrowheads). Examples of nonmyf5-positive TUNEL-positive cells (triangles) and myf5positive nonTUNEL-positive (open arrowheads) cells are also evident. White boxes denote the specific
area of the wound viewed in the subsequent three single-channel images. Representative images taken
were from n = 6 independent fish. DAPI, 4,6-diamidino-2-phenylindole. (G) Quantification of the total
amount of TUNEL-positive cells. Both WT and mutant fish have comparable levels of TUNEL-positive
cells throughout muscle regeneration. Scores shown are means for each measured day postneedlestick injury (DPI) from n = 6 independent fish per condition (mutant or WT). Error bars indicate mean +
SEM. Statistics: t test, two-tailed. NS, not significant. (H) Quantification of the number of myf5expressing cells costained with TUNEL, expressed as a percentage the total number of myf5-positive cells in
the wound site. Scores shown are means for each measured day post injury from n = 6 independent fish
per condition (mutant or WT). Error bars indicate mean + SEM. Statistics: t test, two-tailed. ***P < 0.001.


8 JULY 2016 VOL 353 ISSUE 6295

deeper cells of the myotome; and is not expressed

in the progenitor layer of the ECL (34) (Fig. 3, A to
F, and fig. S8, n = 5 fish). cmet is also prominently
expressed in nonmuscle progenitor populations,
including primary motor neurons and the lateral
line nerve (35) (Fig. 3, A and D, and fig. S8, H to J).
Furthermore, the expression of cmet persists into
adulthood in a fiber-associated niche (fig. S8, A to
G), similar to that described for the mammalian
satellite cell. This observation is consistent with
the regenerative capacity of adult zebrafish muscle,
as well as data from previous studies of adult fiber
isolation in zebrafish (13, 31, 36). To better characterize these cells in vivo, we generated a bacterial
artificial chromosome (BAC) transgenic line that
expressed both KALTA4 and mCherry from the
cmet promoter. This Tg(cmet:mCherry-T2A-KALTA4)modified BAC faithfully recapitulated the expression of the cmet gene upon germline transmission
(Fig. 3, A to F, and fig. S8, A to G). Furthermore,
an antibody against cMet (36) colocalizes with
mCherry expression in the myotome evident within the cmet BAC transgenic line, validating that
this line reports endogenous cmet expression (Fig.
3 and fig. S8, H to J). This transgenic line demonstrated a marked increase in cmet-positive cell
number per myotome section throughout the early
development of the fish, leading up to metamorphosis (up to 15 mm in total length), after which
this population maintained a stable number
throughout postmetamorphic development (15
to 30 mm total length), suggesting a consistent
requirement of this cell type within the myotome
throughout the life span of the fish (fig. S8, K and
L). Using this transgenic line, we established that
cmet and pax3a double-positive and cmet and
pax7a double-positive cells constituted a minor
subset of the interstitial myotomal cells (~15%
of non-ECL, myotomal, pax3- or pax7-positive cells)
(Fig. 3, A to J, n = 3 per transgenic combination).
These cells express a cyclin-dependent kinase inhibitor and therefore probably represent a quiescent population (Fig. 3, K to N, n = 8 fish).
Next, we performed in vivo ablation of the
aforementioned cmet cells to determine whether
these cells are required for regeneration in zebrafish. We crossed the Tg(cmet:mCherry-T2A-KALTA4)
line to Tg(uas:E1b:Eco.NfsB-mCherry), which results
in the expression of the nitroreductase enzyme
specifically within cmet-positive cells. Addition
of metronidazole (Met) to these embryos ablated
cmet-positive cells (Fig. 4, A to D, n = 30 fish) and
resulted in a failure to initiate regeneration and
de novo myogenesis at the wound site (Fig. 4, E
to N, n > 20 per treatment and probe). As cmet
is also expressed within motor neurons, it remained possible that the coablation of neurons
with cmet-positive muscle stem cells could also
directly or indirectly affect the regeneration of
muscle cells that they innervate. To control for
this possibility, we made use of a previously generated transgenic line, Tg(cmetMN:GFP-T2AKALTA4), which contains only a subset of the
cmet regulatory regions evident in the full-length
BAC clone and expresses KALTA4 and GFP only
within the motor neuron expression domain (35).
Crossing this line to Tg(uas:E1b:Eco.NfsB-mCherry) SCIENCE

Downloaded from on July 8, 2016



cells. To assay the effect of these two different

ablation strategies on muscle regeneration, we
performed needle-stab injuries on ablated animals and assayed the regenerative capacity of injured larvae. Although the chemical ablations
carried out in the BAC transgenic line resulted in
defects in muscle regeneration, ablation of cmet-

Fig. 3. cmet expression specifically marks satellite-like cells required for zebrafish muscle regeneration. (A) Maximum-projection confocal images of Tg(pax3a:GFP) (green) and Tg(cmet:mCherry-T2AKALTA4) (red) double-transgenic fish. (B and C) Single confocal sections through the myotome show
pax3a and cmet double-labeled cells [(B) parasagittal, (C) transverse]. (B to B and C to C) Highmagnification views of regions boxed in (B) and (C), respectively (arrowheads, pax3a-high and cmet-high
double-positive cells; arrows, pax3a-high and cmet-lowexpressing cells). (D) Maximum-projection confocal images of Tg(pax7a:GFP) (green) and Tg(cmet:mCherry-T2A-KALTA4) (red) double-transgenic fish.
(E and F) Single confocal sections through the myotome show pax7a and cmet double-labeled cells within
the myotome [(E) parasagittal, (F) transverse]. (E to E and F to F) High-magnification views of regions
boxed in (E) and (F), respectively (arrowheads, double-positive cells; arrows, single-positive cells). (G and
H) Quantification of pax3a-only versus pax3a and cmet double-labeled cells (G) or pax7a-only versus
pax7a and cmet double-labeled cells (H) per myotome. (I and J) Quantification of pax3a and cmet
double-positive cells as a percentage of total pax3a-positive cells (I) or pax7a and cmet double-positive
cells as a percentage of total pax7a-positive cells (J). For both pax3a and pax7a, cmet colabels 10 to 15%
of the total population. Error bars indicate mean + SEM. (A to F) Representative images were selected
from n = 6 independent fish. (G to J) Numbers are derived from total myotomal counts of three
individual larvae per transgene combination. (K to N) Cross section of 4-dpf [(K) to (K)] and 10-mm
[(L) to (N)] Tg(cmet:mCherry-T2A-KALTA4) larvae costaining with an antibody against the cyclindependent kinase inhibitor ink4b. Arrows highlight colocalizing ink4b nuclear staining that occurs in
cmet-positive cells (representative images from n = 8 sectioned embryos).

positive motor neurons had no effect on the ability

of larvae to regenerate (Fig. 4, Q to U). These
studies collectively define cmet-positive cells as
the functional equivalent of mammalian satellite
cells, specifically required for muscle regeneration
within the zebrafish myotome.
Asymmetric divisions generate clonally
restricted progenitors
To examine how the cmet-positive cell population
is activated during injury, we undertook continuous
time-lapse analyses of the cmet-positive cells in
several transgenic backgrounds that had undergone
laser ablation injury. Analyses within injured
animals double-transgenic for Tg(myf5:GFP) and
Tg(cmet:mCherry-T2A-KALTA4) revealed that cmet
and myf5 double-positive cells are associated with
the injury site and appear as doublets within cmetonly cells (Fig. 5, A to D, and fig. S9E, n = 6 timelapse analyses of injured larvae, n = 35 doublets
identified). These results suggest that cmet-positive
cells are activated by injury to divide asymmetrically
to produce a myf5-positive progenitor population
that, in turn, generates the de novo myogenesis
evident during muscle regeneration described
above. Collectively, our results suggest that the
myf5-positive cells observed in earlier experiments
contained in this paper (Figs. 1H and 2A and
movies S4 and S5) are, in fact, cmet-low/myf5high daughter cells that are responding to signals
from the injury site and from regenerating myofibers. Next, we undertook time-lapse imaging
within injured animals double-transgenic for
Tg(pax7a:GFP) and Tg(cmet:mCherry-T2A-KALTA4)
to capture the earliest division of cmet-positive
cells. cmet and pax7a double-positive cells migrate
to the wound site and proliferate, indicating that
cmet marks injury-responding cells during muscle
regeneration. Division of activated cmet and pax7a
cells was asymmetric, such that after each division
a rounded cmet-high/pax7a-lowexpressing cell
was maintained and a pax7a-high or cmet-low
expressing progenitor population was created (Fig.
5E, movies S6 and S7, and fig. S9). Individual cmetpositive cells were identified that gave rise to
multiple cmet-low and pax7a-highexpressing progeny (n = 6 time lapses of injured larvae, n = 38
doublets identified; fig. S9 and movies S6 and S7).
Furthermore, our analyses indicate that pax7 cells
are proliferative within the wound site at 2 dpi,
suggesting that cmet-low/pax7a-high cells proliferate, elongate, and undergo differentiation in
the manner we have described above (Fig. 5E;
movie S6; and fig. S6, D and H). Collectively, these
observations suggest that the cmet-high/pax7alow cells evident within our time-lapse analyses
represent a regeneration-specific stem cell compartment that divides asymmetrically to self-renew the
stem cell compartment and produce proliferative
daughters cells required to effect regeneration.
In vitro studies have suggested that satellite
cells divide in an apical-basal fashion relative to
the plane of the muscle fiber, a process that subsequently dictates the fate of daughter cells (5).
More recent analyses have defined a role for the
dystrophin-associated glycoprotein complex in the
control of this process whereby dystrophin localization
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016

results in the expression of the nitroreductase

enzyme specifically within cmet-positive motor
neurons. Addition of Met to these embryos resulted
in complete ablation of primary motor neurons
(Fig. 4, O to P). In parallel, we also performed
ablations in the cmet BAC line, which ablates both
cmet-positive motor neurons and muscle stem


this asymmetry appears random relative to muscle fiber orientation (fig. S8, M to N), which suggests that the initial asymmetric cell divisions of
the activated stem cell would likewise be in multiple orientations. This suggests that local micro
niches could provide specific extracellular matrix
cues that dictate the localization and polarity of
the stem cell divisions initially.
Given the low number of these cells evident in
the larval myotome (~15% of pax7a- or pax3apositive cells in the myotome; Fig. 3, I and J), the

Fig. 4. cmet-positive cells are required for regeneration.

(A to C) Maximum-projection confocal images of larvae
transgenic for Tg(pax7a:GFP); Tg(cmet:KALTA4-T2AmCherry); Tg(uas:E1b:Eco.NfsB-mCherry), which results
in the expression of the nitroreductase enzyme specifically within cmet-positive cells. Addition of the prodrug
metronidazole (Met) to these triple-transgenic larvae
[(B) and (C)] results in ablation of cmet-expressing cells
[arrowheads in (A)] but does not affect myoblasts expressing pax7a alone. The images from the cmet ablation are quantified in (D), indicating that 1 mM of Met is
sufficient to induce significant reduction in cmet-positive
cell numbers. mCherry-positive cellular debris collects
superficially in metronidazole-treated larvae [(B) and
(C)], which results not only from myoblast cell death
but also degradation of the lateral line [asterisks in
(A)] and motor neurons [arrows mark cell bodies in
(A)], which also express cmet. The appearance of this
mCherry-positive cellular debris is used as a positive
control to show that the ablation has worked efficiently.
Error bars in (D) indicate mean + SEM. ***P < 0.001. (E to
N) Ablation of cmet-positive cells through metronidazole
addition [(K) to (M)] results in a failure to initiate de
novo myogenesis, which is indicated as failure to initiate
the expression of myogenic regulatory factor gene expression at the wound site. Such a failure does not occur in
control animals that have not received metronidazole [-Met,
(H) to (J)] or those larvae that do receive metronidazole
(+Met) but do not carry the Tg(uas:E1b:Eco.NfsB-mCherry)
transgene and therefore do not express nitroreductase
(-nfsb) [(E) to (G)]. Representative images of n > 20
animals per treatment group and probe. Insets show
high-magnification views. (N) Quantitation of myogenic
gene expression of ablated and nonablated larvae. Larvae were placed into arbitrary bins of expression (high,
medium, low, or none), examples of which are shown in
(E) to (M). (O to U) Ablation of cmet-positive motor neurons inhibits regeneration. (O) Transgenic line Tg(cmetMN:
GFP-T2A-KALTA4) containing a subset of the cmetregulatory regions evident in the full-length BAC clone
expresses KALTA4 only within the motor neuronexpression
domain. Crossing this line to Tg(uas:E1b:Eco.NfsB-mCherry)
results in the expression of the nitroreductase enzyme (red) specifically within cmet-positive motor
neurons. (P) Addition of Met to these embryos results
in complete ablation of primary motor axons (positions
marked by arrows) and their arbors, leaving only
nonspecific mCherry cellular debris after ablation. NT,
neural tube. (Q to U) Larval muscle regeneration is not
impaired in motor neuronablated animals [+Met +injury, (S)] and is
identical to injured nonablated controls [-Met +injury, (R)]. By contrast,
ablation of both motor neurons and cmet-positive muscle stem cells in the
cmet BAC transgenic line impairs regeneration (+Met +injury). (Q to T)
Representative images of larvae under indicated conditions imaged for
birefringence under polarized light. (U) Quantitation of birefringence


8 JULY 2016 VOL 353 ISSUE 6295

asymmetric model of stem cell division indicates

that a finite number of regeneration-specific stem
cells, represented by cmet expression, are activated
after injury, and that these cells would generate a
limited set of clonally related lineages to effect
repair. Clonality can also be predicted from several
of the current models of satellite cell action in
amniote muscle repair that propose the existence
of a small subset of satellite cells with true stem
cells properties (79, 14). Specifically, the asymmetric model suggests that particular satellite

Downloaded from on July 8, 2016

within the stem cell niche interacts with the PAR

complex to regulate asymmetric division of the
satellite cell (37). To determine whether a similar
mechanism could control asymmetric division
seen in our in vivo model, we used a gene-trap line
driving the citrine fluorophore under the native
dystrophin locus, Gt(dmd-citrine) (38), in combination with our Tg(cmet:mCherry-T2A-KALTA4)
line. Examination of quiescent satellite cells showed
us that dystrophin is specifically expressed on cmetpositive cells and is localized asymmetrically, but

levels relative to uninjured controls reveals that motor neuronablated

larvae regenerate similarly to nonablated controls. However, ablation of
both motor neurons and cmet-positive muscle stem cells in the cmet
BAC transgenic line results in significantly lower birefringence, which
points to impairment of muscle regeneration. Error bars indicate mean +
SEM. Statistics: t test, two-tailed. ****P < 0.0001; ***P < 0.001; **P < 0.01. SCIENCE

cells maintain the ability to return to quiescence

via asymmetric DNA segregation followed by
differential gene expression, cell cycle entry, and
metabolic status. Thus, this model indicates that
all myogenic cells generated during repair would
be clonally related to rare stem satellite cells
(79, 14). However, this clonal model of stem cell
action lacks in vivo validation. To provide evidence
for this mode of satellite cell action and validate
our own observations, we crossed the Tg(cmet:
mcherry-T2A-KALTA4) line with the ubiquitously
expressed Tg(ubi:zebrabow) line (Fig. 5, F to H, n =
22 fish) (39). These double-transgenic embryos
were further injected with a UAS:CreERT2 construct, which is responsive to the KALTA4 promoter. We then performed needle-stick injuries
on these animals. Two outcomes were possible
from this experiment. If cmet expression simply
reflects a progenitor phase of muscle proliferation, then Cre-activated deletion in cmet-positive
cells would result in multiple deletions within independently proliferating cells and, hence, a wide
range of distinct zebrabow rearrangements leading
to many different colored myoblast lineages. However, if cmet expression denotes a stem population
that generates a clonally related progenitor compartment, Cre-mediated deletion would generate
rearrangements within a limited number of stem
cells. In turn, these stem cells would generate the
bulk of proliferating myoblasts in the regenerate,
with progeny of individual zebrabow-recombined
stem cells all expressing an identical color upon
differentiation into muscle cells. Upon needlestick injury of the Tg(cmet:mcherry-T2A-KALTA4),
Tg(ubi:zebrabow) double-transgenic line into
which UAS:CreERT2 was injected, only a very few
distinct colors are evident that cluster in homogeneously expressed and regionally localized clones
of myoblasts and muscle cells produced during
regeneration (Fig. 5, F to H, n = 22 fish). This result indicates that the cmet-positive cells represent
the injury-responsive clonal stem cell compartment of the zebrafish myotome (fig. S10J).
To more broadly assay the clonal nature of muscle regeneration, independent of the cmet stem cell
compartment, we performed needle-stick muscle
injuries in animals that were doubly transgenic
for Tg(ubi:zebrabow) and Tg(msgn1:CreERT2).
The addition of tamoxifen to these fish during
embryogenesis results in zebrabow rearrangements in all cells derived from the somites, as
the mesogenin (msgn1) promoter is somite-specific
(40). Because the somite is the embryonic compartment from which all muscle stem and progenitor
cells derive, this combination of transgenes enables
the generation of a musclebow fish, allowing the
clonal relationships of muscle cells to be assessed
(Fig. 5, J to Q; fig. S10, A to I, n = 13 fish; and movie
S8) (39). Tamoxifen addition to uninjured embryos
early during embryogenesis results in the labeling
of muscle fibers within the embryo and reveals
random deletions of the zebrabow array, leading
to individual fibers expressing unique combinations of fluorophores (Fig. 5, I to M; fig. S10, A to I;
and movies S8 and S10, n = 13 fish). No bias in
any specific zebrabow deletion is detected by
this analysis, with individual fibers displaying ranSCIENCE

dom spectral values upon differentiation (Fig. 5, I

to M; fig. S10, A to I; and movies S8 and S10).
Therefore, injury in these animals allows for global
assay of the number of somite-derived lineages
that contribute to zebrafish muscle regeneration.
To statistically evaluate the clonal nature of regeneration, we manually determined the hue (radians) and saturation (percentage) values for each
fiber and plotted them such that each coordinate
represents the color identity of a particular fiber.
We then calculated the distances between all
coordinates to provide a statistically evaluated
measure of clonality (Fig. 5I; see methods for computational methodology). Our analysis indicated
that, in stark contrast to muscle fibers that result
from growth, regenerated muscle fibers are highly
clonal, as a limited numbers of recombined fluorescent phenotypes contribute to regeneration, consistent with a stem cell origin for these cells (Fig.
5, N to Q; fig. S10, A to I; and movies S9 and S10).
However, multiple clones, contributing individual
single clonally restricted lineages, can contribute
to regeneration of larger wounds, eliminating the
possibility that a single clonal lineage marked
early in development generates all satellite cells
in zebrafish (e.g., fig. S10, E, E, I, and I). Furthermore, our analysis confirmed that the stem cells
that contribute to regeneration derive from the
somite, as these cells exhibit a recombined fluorescent phenotype generated from the expression
of the somite-specific meosgenin Cre line. Collectively, our analyses, combined with previously published results (79, 14), indicate that vertebrate
muscle regeneration results from the activation
of limited set of somite-derived stem cells that
generate a clonally restricted proliferative myoblast population, which in turn differentiates to
effect muscle repair.
We analyzed the morphological, cellular, and
genetic basis for muscle regeneration in zebraish.
Using transgenes that mark muscle stem and
progenitor cells, in combination with long-term
imaging modalities, we imaged the entire process
of regeneration, from the initial injury to the
formation of newly differentiated muscle fibers at
the wound site in a vertebrate. We show that a stem
cell niche equivalent to the mammalian satellite
cell system operates within the zebrafish myotome. This observation suggests that the satellite cell is part of an evolutionary ancient stem
cell system that is present throughout the vertebrate phylogeny and may well be a basal feature
of gnathostomes.
Our time-lapse analyses enabled us to characterize a number of stereotypical phases of repair
involving specific cell cohorts (Fig. 6 and fig. S10J).
Although many of these morphogenetic events are
canonical ones, in terms of known stem celldriven
regenerative processes (activation, proliferation of
a progenitor population, and differentiation), the
dynamic nature of the morphogenetic changes of
both the activated stem cell and the proliferating
progenitor population is surprising. Although migration of activated satellite cells on fibers has been
documented in vitro (41), there has been no indi-

cation from such paradigms that the migration of

activated muscle stem cells would deploy such a
polarized phenotype. The polarized shape that
myf5-positive cells adopt to migrate to the site of
injury is therefore suggestive of a novel directional
migratory mode. A recent examination of the
role of activated pax7-positive cells during in
vivo muscle repair in the mouse revealed similar
polarized phenotypes for activated migrating murine myogenic cells (10), indicating the potential
relevance of our observation to mammalian muscle biology. Due to current limitations of examining this process in vivo, little other information
exists as to the mechanism used by migrating
activated satellite cells to track to the wound site
in amniotes. These limitations are largely overcome by use of the zebrafish larval system.
Furthermore, the interplay between the proliferating myogenic cells that enter the wound site
with injured fibers in the initial contact phase of
regeneration and secondarily with the undamaged
remaining fibers at the injury site was also unexpected. A lack of information is available about
how the injury site itself acts to direct the architecture and behavior of the regenerating fibers
during skeletal muscle regeneration, as existing
studies have almost exclusively focused on the
factors that influence the stem cell compartment.
However, the importance of the wound environment
has been emphasized in recent in vivo investigations performed during early muscle repair in the
mouse, which showed that myogenic progenitors
at the injury site use the extracellular matrix left
from dead and damaged muscle fibers to provide
orientation cues for cell division and alignment
(10). In this regard, the lassoing event that appears
to guide differentiating cells at the wound by
the surrounding uninjured fibers was an extension of these observations. It suggests that instead
of being passive bystanders in the regenerative
process, uninjured fibers themselves may play
a role in directing differentiating progenitors to
regions of the wound that are most in need of
new fiber addition. It also suggests that the kinetics of repair at the wound, and even aspects of
myocyte differentiation, may partly be dictated
by the nature and availability of intact fibers to
participate in this process.
Using the cmet and pax7 transgenes as markers
of the stem cell population, we have been able to
examine activated cells immediately after injury
by continuous time-lapse analyses. Our analyses
detected the existence of two clonally related sets
of cells during regeneration: a cmet-high/pax7-low
stem cell population and a pax7-high/cmet-low
progenitor population. Time-lapse analyses revealed that the cmet-high population divided
asymmetrically to generate the differentiationcompetent cmet-low population, as well as a cmethigh cell that maintained the ability of producing
both cmet-low proliferative daughter cells and a
self-renewing cmet-high undifferentiated stem
cell. Previous analyses of murine single-fiber
explants have suggested that the plane of division
of a satellite cell, relative to the underlying basal
lamina, dictates the fate of individual satellite
cell daughters (5). Specifically, division of a satellite
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 5. Zebrafish satellite-like cells represent a

lineage-restricted injury-responsive stem cell that
undergoes asymmetric division during regeneration. (A to D) Maximum-intensity-projection confocal
images of Tg(myf5:GFP) (green) and Tg(cmet:mCherryT2A-KALTA4) (red) double-transgenic fish 2 days after
needle injury. Double-positive cells (arrows), cells with
Tg(myf5:GFP) alone (asterisks), and cells with Tg(cmet:
mCherry-T2A-KALTA4) alone (arrowheads) are marked.
Insets reveal the existence of cells with Tg(cmet:mCherryT2A-KALTA4) alone and double-positive doublet cells
associated with the injury site, illustrated in the region
framed in the combined differential interference contrast image in (D). Representative images were taken
from n = 6 independent fish. (E) Asymmetric division
generates a cmet-high/pax7-low stem cell compartment
and a cmet-low/pax7-high differentiation-competent
progenitor population after laser ablation injury. Continuous confocal time-lapse imaging of 4-dpf larvae
immediately after laser injury shows cells transgenic
for Tg(cmet:mCherry-T2A-KALTA4) (red) and Tg(pax7a:
GFP) (green), as well as those imaged in the brightfield
channel (insets without brightfield). A single cmet-high/
pax7-low cell (arrow) maintains cmet expression through
several rounds of division that generate cmet-low/pax7high cells (arrowheads), which subsequently elongate
and differentiate.Time (t) is given in minutes. Scale bar,
50 mm. (F to H) Needle-stick injury of larvae transgenic for Tg(ubi:zebrabow) and Tg(cmet:mCherry-T2AKALTA4) injected with a UAS:CreERT2 DNA construct.
Tamoxifen addition results in limited recombination
events generating a highly restricted set of colors in
regenerating fibers and myoblasts. Red is the default
expression color that appears in every cell; the
expression of any other fluorophore represents a
specific recombination. Only two regenerating lineages
are evident: one from a recombination of the zebrabow
cassette that generates blue fibers [marked by
arrows in (G)] and myoblasts and a second that
generates pink fibers [marked by arrowheads in
(G)] and myoblasts. (G) High-magnification view of
the region boxed in (F). (H) Cross section at the level
marked in (G). Representative images are from n = 14
imaged larvae. Scale bar in (F), 50 mM. (I to Q) Injured
larvae transgenic for Tg(msgn1:CreERT2) and Tg(ubi:
zebrabow), to which tamoxifen addition results in
clonal labeling of all somite-derived cells. (I) Quantitation of the clonality of muscle regeneration. To statistically evaluate the clonal nature of regeneration, the
hue (radians) and saturation (percentage) values were
manually determined for each fiber and plotted such
that each coordinate represents the color identity of
a particular fiber. Distances between all coordinates
were then calculated to provide a statistically evaluated measure of clonality. Error bars indicate mean +
SEM. Statistics: t test, two-tailed. ***P < 0.001. (J to M) Confocal sections
through an uninjured larvae at 4 dpf, revealing the specific differentiated
muscle fibers clonally marked by individual zebrabow rearrangements. (N to
Q) The same larvae as in (J) to (M) (at 3 dpi and 7 dpf), revealing that
regenerating fibers all express the same zebrabow green rearrangement
(regenerated fibers bounded by white line), which indicates that regenerat-

cell in the apical-basal plane relative to the

muscle fiber generates an apically positioned,
differentiation-competent progenitor and a basally located self-renewing stem cell. A more recent

8 JULY 2016 VOL 353 ISSUE 6295

Downloaded from on July 8, 2016


ing fibers are clonally related and derive from a single stem cell {sagittal [(J)
and (N)], coronal [(L) and (P)], and transverse [(K) and (O)] sections at the
levels indicated}. Scale bars, 50 mm. Representative images were taken
from n = 13 independent fish. (M and Q) Plotted spectral values (dots) of
individual fibers from the same larvae within uninjured (M) and regenerated (Q)
muscle, revealing the comparative clonality of regenerating fibers after injury.

imaging study of in vivo mouse muscle repair has

suggested that activated myogenic progenitors at
the wound site divide in a planar fashion, although
this study was unable to assay the process of sat-

ellite cell activation (10). Although the highly migratory nature of the activated satellite cells in
vivo precludes a definitive assignment of relative
polarity of division, our observations suggest SCIENCE


that the initial division of activated satellite cells

probably occurs in a random polarity with respect to the plane of the muscle fiber in zebrafish.
Furthermore, the clonality of regeneration we
observe in response to muscle injury is a major
finding of this work. Clonality is a direct prediction of the self-renewing immortal stem cell
hypothesis generated from seminal in vitro observations (59), and our work provides direct
in vivo validation of this hypothesis. We conclude that asymmetric divisions occur during
muscle regeneration to generate the clonally
related progenitors required for muscle repair,
resolving a long-term debate surrounding the
existence of this mechanism of stem cell selfrenewal and muscle repair in vivo.
Materials and methods
Zebrafish strains and maintenance
Wild-type (WT) and transgenic lines Tg(-80.0myf5:
eGFP) [referred to as Tg(myf5:GFP)] (42), Tg(actc1b:
mCherry) (43), Tg(actc1b:BFP) (43), Tg(actc1b:
mCherryCAAX) (44) ,Tg(smyhc1:GFP) (45), TgBAC
(pax7a:GFP) [referred to as Tg(pax7:GFP)] (12),
TgBAC(pax3a:GFP) [referred to as Tg(pax3:GFP)]
(12), Tg(ubi:zebrabow) (39), Tg(msgn1:CreERT2)
(40), Tg(UAS:nsfB-mCherry) (46), Tg(cmetMN:
GFP-T2A-KALTA4) (34), Gt(dmd-citrine)ct90a

[(referred to as Gt(dmd-citrine)] (38), Tg(actc1b:

Cre), and Tg(ef1a1:LOXP-eGFP-LOXP-mCherry)
were maintained on TE WT background. Staging
and husbandry were performed as previously
described (47). A new transgenic line produced
by microinjection into TE WT fish according to
standard procedures (48) and used in this study
is TgBAC_CH211-105D03(cmet:mCherry-T2AKALTA4) [hereafter referred to as Tg(cmet:mCherryT2A-KALTA4)], a BAC construct. Mutant strains
used were myf5hu2022 (28), myf6hu2041 (28), myog fh265
(27), and myodhu2024 (49), all maintained on AB
background. These mutants were genotyped by
sequencing of polymerase chain reaction products amplified from fin clip or embryo genomic
5-CTGTGCGCAGAGTCAAAG-3 for myf6, and
Gateway cloning and mosaic transgenesis
Gateway cloning was performed using the Gateway Tol2kit to create transgenic constructs, according to standard Invitrogen protocols (50). The

UAS:CreERT2 and actc1b:nlsGFP constructs were

created using the Gateway method. To generate
mosaically transgenic zebrafish, individual plasmid DNAcontaining constructs flanked by tol2
sites were delivered by microinjection into embryos at the single-cell stage, directly into the
cell. The plasmid DNA concentration was typically at 25 ng/ml. The Tg(actc1b:Cre) and Tg(ef1a1:
LOXP-eGFP-LOXP-mCherry) transgenic fish were
generated from constructs also made using
the Gateway method. To generate stable transgenic zebrafish, a mixture of mRNA-encoding
tol2 recombinase and a DNA plasmid containing a construct flanked by tol2 sites was
delivered by microinjection directly into the cell
at the one-cell stage of development. mRNA was
delivered at 20 ng/ml and the DNA construct at
25 ng/ml.
Needle-stick injury
For needle-stick injury induction, tricaineanaesthetized 4-dpf larval zebrafish were placed
on a 24.5 mmby76.2 mm glass slide in a drop
of water and maneuvered into a lateral lying position with their heads pointing toward the left. A
single needle-stick injury into the dorsal epaxial
musculature above the cloaca was performed at
an angle of 75, using a 30-gauge needle. Subsequently, fish were immediately transferred to
fresh embryo water (E3) to recover. The ventral
side of the fish relative to the site of injury was
used as an uninjured internal control.
Laser injury
For laser-induced injury, tricaine-anaesthetized
4-dpf larval zebrafish were placed in grooves on a
specialized silicon block, similar to the one used
for microinjection of embryos, mounted in a petri
dish. A thin layer of lowmelting point agarose
[0.5% weight/volume agarose in E3 (Invitrogen)],
maintained in a molten state at 36C, was applied
to each fish. As soon as the lowmelting point
agarose was set, the petri dish was filled with
E3 containing tricaine to maintain the anaesthetized state throughout the injury procedure and
for subsequent time-lapse photomicroscopy. A
Micropoint laser connected to a Zeiss Axioplan
microscope was used for injuring, as previously
described (51), with a 40X water immersion objective and a laser pulse at a wavelength of 435 nm
for cell ablation. Time-lapse recordings were generated on a Zeiss LSM 710 Live Duo confocal microscope (Carl Zeiss Microimaging).
Metronidazole ablation experiments
Cell ablation using this system was performed
as previously described (52), with some modifications. Briefly, the prodrug metronidazole
(Sigma) was dissolved in E3 water as a 5 or
1 mM working solution, as indicated, along with
added 0.2% dimethyl sulfoxide and 0.003% 1phenyl-2-thiourea in 10% Hanks saline (PTU).
Transgenic zebrafish were incubated in this solution immediately after injury (4 dpf) at 28.5C,
solutions were changed every day, and fish were
euthanized and fixed at 7 dpf for whole-mount
in situ hybridization (WISH). Control fish were
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016

Fig. 6. A model for in vivo muscle regeneration. (A to J) In vivo

repair of muscle from a defined
stem population occurs through a
number of stereotypical phases.
The initial phase (migration and
contact) is characterized by
migration of the satellite-like cell
populations into the wound area.
Once in the wound, cells immediately associate as rounded cells
with severed and dying fibers and
remain in contact for several
hours. In the second phase (proliferation and activation), activated
stem cells undergo an asymmetric
division to generate the proliferating compartment that undergoes
further division. In the next phase
(bipolar extension and interaction), proliferating cells extend to a
bipolar shape and lose contact
with the dying cells around them.
Bipolar cells in the wound then
actively migrate toward each other,
extend filopodia to make contact, and
then recoil and move rapidly apart
from one another. Next, resident
uninjured differentiated muscle cells
extend filopodial-like membrane protrusions, which adhere and lasso the
activated cells. Filopodial retraction
guides the captured cell to align next
to the differentiated muscle cell from
which the filopodia extends. Cells
then elongate and differentiate into
muscle fibers, marking the end of the final phase (resident fiber capture and differentiation). Within 7 days,
muscle fiber repair is complete, and the injury site is indistinguishable from surrounding uninjured fibers.

subjected to the same injury and treatment procedure without metronidazole.

WISH and immunohistochemistry
In situ mRNA hybridization and immunohistochemistry were performed as previously described
(31). The digoxigenin-tagged probes used were
pax7a (53), cmet (31), myf5 (54), mrf4 (55), and myod
and myog (56). Analyses of expression levels after in
situ hybridization involved randomly mounting
stained embryos, which were blinded to treatment
group. Larvae were photographed and scored into
arbitrary bins of none, low, medium, and
high, on the basis of the experimental spread of
expression. Images representative of the overall
phenotype were then presented in the data.
Primary antibodies used were rabbit anti-GFP
[1:500 (Invitrogen)], mouse anti-Pax7 [1:10 (Developmental Studies Hybridoma Bank, DSHB)],
mouse F59 antislow myosin heavy chain [1:10
(DSHB)], and rabbit anti-phospho-histone-H3
[1:1000 (Sigma)]. The p16 INK antibody (1:4) (57)
probably recognizes proteins that are produced
from a single locus ancestral to mammalian
CDKN2A and -2B, as only a single syntenic gene
is present in zebrafish. Hence, this antibody could
recognize a range of potential cyclin inhibitors.
Secondary antibodies used include goat antirabbit Alexa Fluor 488 [1:1000 (Invitrogen)], goat
anti-rabbit Alexa Fluor 568 [1:1000 (Invitrogen)],
goat anti-mouse Alexa Fluor 488 [1:1000 (Invitrogen)], goat anti-mouse Alexa Fluor 568 [1:1000,
EdU labeling
Cell proliferation was determined by 5-ethynyl-2deoxyuridine EdU labeling according to established
conditions (58), using 7 mM EdU (Invitrogen) dissolved in E3 medium. Briefly, larval fish were incubated for a 48-hour block before fixation. Fish
were euthanized, fixed, and subsequently processed
using the Click-iT EdU Imaging Kit (Invitrogen).
This processing occurred after antibody staining,
using the Click-iT solution.
TUNEL staining
Cell apoptosis was determined by terminal deoxynucleotidyl transferasemediated deoxyuridine
triphosphate biotin nick end labeling (TUNEL)
staining, performed on whole larvae as previously
described (52), with some modifications. To stain
for apoptosis, a TMR Red In Situ Cell Detection
Kit (Roche) was used according to manufacturers
specifications. Samples were incubated in stain
solution in the dark for 2 hours at 37C before
processing and observation by fluorescence and
confocal microscopy.
Birefringence imaging
Birefringence is the property of intact muscle tissue
to diffract polarized light (damaged muscle tissue
lacks this quality). Birefringence was measured
as previously described (59). Briefly, anaesthetized zebrafish were placed on the Leica DMIRB
(Leica Microsystems) microscope stage in a glassbottom FluoroDish (World Precision Instruments).
The exposure levels of the attached polarizing

8 JULY 2016 VOL 353 ISSUE 6295

filters were automatically adjusted by the integrated Abrio Software [CRI (Hinds Instruments)],
and the resulting final picture shows an averaged
polarizing effect. The resultant birefringence images
were analyzed for birefringence densitometry over
the required area using the software FIJI.
Zebrabow analyses
Tamoxifen treatments were conducted as previously described for the Tg(ubi:zebrabow), Tg(msgn1:
CreERT2) double-transgenic larvae (40). At 4 dpf,
the fish were imaged and subsequently injured
by needle stick. At 7 dpf, fish were reimaged at
the injury site. Optical transverse cross sections
were then generated from confocal stacks with
Imaris (Bitplane), and color profiles of regenerated fibers at the injured site were measured
in Photoshop (Adobe) as previously described
(39). For clonal analysis, images were processed
as described in (39), with modifications. Images
were opened in Photoshop CS5.1, and the eyedropper tool (sample size set at 5 by 5 pixels)
was used to measure the fiber of interest. Color
display mode was set to hue saturation brightness, and the hue and saturation values in the
color window were manually recorded. Hue
(radians) and saturation (percentage) values
were converted into x-y coordinates to be plotted
on a Cartesian graph using basic trigonometry
(hue, degrees; saturation, hypotenuse; adjacent or opposite side, point of interest). Each
coordinate represented the color identity of a
particular fiber. Distances between all clones
were then calculated with following formula:
Distance x1 x2 2 y1 y2 2 . Distance
values were then imported into Prism software
to conduct a Students t test. For the Tg(cmet:
mCherry-T2A-KALTA4); (UAS:CreERT2); Tg(ubi:
zebrabow) transgenic larvae, tamoxifen was added
at 4 dpf, immediately before injury, and samples
were incubated postinjury for a further 8 hours
before washing into embryo media.

1. P. Seale et al., Pax7 is required for the specification of

myogenic satellite cells. Cell 102, 777786 (2000).
doi: 10.1016/S0092-8674(00)00066-0; pmid: 11030621
2. D. D. Cornelison, B. J. Wold, Single-cell analysis of regulatory
gene expression in quiescent and activated mouse skeletal
muscle satellite cells. Dev. Biol. 191, 270283 (1997).
doi: 10.1006/dbio.1997.8721; pmid: 9398440
3. C. Lepper, T. A. Partridge, C. M. Fan, An absolute requirement
for Pax7-positive satellite cells in acute injury-induced skeletal
muscle regeneration. Development 138, 36393646 (2011).
doi: 10.1242/dev.067595; pmid: 21828092
4. C. A. Collins et al., Stem cell function, self-renewal, and
behavioral heterogeneity of cells from the adult muscle
satellite cell niche. Cell 122, 289301 (2005). doi: 10.1016/
j.cell.2005.05.010; pmid: 16051152
5. S. Kuang, K. Kuroda, F. Le Grand, M. A. Rudnicki, Asymmetric
self-renewal and commitment of satellite stem cells in muscle.
Cell 129, 9991010 (2007). doi: 10.1016/j.cell.2007.03.044;
pmid: 17540178
6. I. M. Conboy, T. A. Rando, The regulation of Notch signaling
controls satellite cell activation and cell fate determination in
postnatal myogenesis. Dev. Cell 3, 397409 (2002).
doi: 10.1016/S1534-5807(02)00254-X; pmid: 12361602
7. V. Shinin, B. Gayraud-Morel, D. Goms, S. Tajbakhsh,
Asymmetric division and cosegregation of template DNA
strands in adult muscle satellite cells. Nat. Cell Biol. 8,
677687 (2006). doi: 10.1038/ncb1425; pmid: 16799552

8. P. Rocheteau, B. Gayraud-Morel, I. Siegl-Cachedenier,

M. A. Blasco, S. Tajbakhsh, A subpopulation of adult skeletal
muscle stem cells retains all template DNA strands after cell
division. Cell 148, 112125 (2012). doi: 10.1016/j.cell.2011.11.049;
pmid: 22265406
9. A. Troy et al., Coordination of satellite cell activation and selfrenewal by Par-complex-dependent asymmetric activation of
p38a/b MAPK. Cell Stem Cell 11, 541553 (2012). doi: 10.1016/
j.stem.2012.05.025; pmid: 23040480
10. M. T. Webster, U. Manor, J. Lippincott-Schwartz, C. M. Fan,
Intravital imaging reveals ghost fibers as architectural units
guiding myogenic progenitors during regeneration. Cell Stem Cell
18, 243252 (2016). doi: 10.1016/j.stem.2015.11.005;
pmid: 26686466
11. S. Knappe, P. S. Zammit, R. D. Knight, A population of Pax7expressing muscle progenitor cells show differential responses
to muscle injury dependent on developmental stage and injury
extent. Front. Aging Neurosci. 7, 161 (2015). pmid: 26379543
12. C. Seger et al., Analysis of Pax7 expressing myogenic cells in
zebrafish muscle development, injury, and models of disease.
Dev. Dyn. 240, 24402451 (2011). doi: 10.1002/dvdy.22745;
pmid: 21954137
13. A. Rowlerson, G. Radaelli, F. Mascarello, A. Veggetti,
Regeneration of skeletal muscle in two teleost fish: Sparus
aurata and Brachydanio rerio. Cell Tissue Res. 289, 311322
(1997). doi: 10.1007/s004410050878; pmid: 9211834
14. D. Montarras, A. Lhonor, M. Buckingham, Lying low but ready
for action: The quiescent muscle satellite cell. FEBS J. 280,
40364050 (2013). doi: 10.1111/febs.12372; pmid: 23735050
15. R. J. Bryson-Richardson et al., Myosin heavy chain expression
in zebrafish and slow muscle composition. Dev. Dyn. 233,
10181022 (2005). doi: 10.1002/dvdy.20380; pmid: 15830374
16. S. H. Devoto, E. Melanon, J. S. Eisen, M. Westerfield,
Identification of separate slow and fast muscle precursor cells
in vivo, prior to somite formation. Development 122,
33713380 (1996). pmid: 8951054
17. R. Matsuda, D. H. Spector, R. C. Strohman, Regenerating adult
chicken skeletal muscle and satellite cell cultures express
embryonic patterns of myosin and tropomyosin isoforms.
Dev. Biol. 100, 478488 (1983). doi: 10.1016/0012-1606(83)
90240-3; pmid: 6653881
18. U. Carraro, L. Dalla Libera, C. Catani, Myosin light and heavy
chains in muscle regenerating in absence of the nerve:
Transient appearance of the embryonic light chain.
Exp. Neurol. 79, 106117 (1983). doi: 10.1016/0014-4886(83)
90382-5; pmid: 6822248
19. S. Sartore, L. Gorza, S. Schiaffino, Fetal myosin heavy chains in
regenerating muscle. Nature 298, 294296 (1982).
doi: 10.1038/298294a0; pmid: 7045696
20. S. Schiaffino, A. C. Rossi, V. Smerdu, L. A. Leinwand,
C. Reggiani, Developmental myosins: Expression patterns and
functional significance. Skelet. Muscle 5, 22 (2015).
doi: 10.1186/s13395-015-0046-6; pmid: 26180627
21. D. T. Burnette, A. W. Schaefer, L. Ji, G. Danuser, P. Forscher,
Filopodial actin bundles are not necessary for microtubule
advance into the peripheral domain of Aplysia neuronal growth
cones. Nat. Cell Biol. 9, 13601369 (2007). doi: 10.1038/
ncb1655; pmid: 18026092
22. E. W. Dent et al., Filopodia are required for cortical neurite
initiation. Nat. Cell Biol. 9, 13471359 (2007). doi: 10.1038/
ncb1654; pmid: 18026093
23. J. C. Fierro-Gonzlez, M. D. White, J. C. Silva, N. Plachta,
Cadherin-dependent filopodia control preimplantation embryo
compaction. Nat. Cell Biol. 15, 14241433 (2013). doi: 10.1038/
ncb2875; pmid: 24270889
24. E. Stanganello et al., Filopodia-based Wnt transport during
vertebrate tissue patterning. Nat. Commun. 6, 5846 (2015).
doi: 10.1038/ncomms6846; pmid: 25556612
25. L. A. Megeney, B. Kablar, K. Garrett, J. E. Anderson,
M. A. Rudnicki, MyoD is required for myogenic stem cell
function in adult skeletal muscle. Genes Dev. 10, 11731183
(1996). doi: 10.1101/gad.10.10.1173; pmid: 8675005
26. B. Gayraud-Morel et al., A role for the myogenic determination
gene Myf5 in adult regenerative myogenesis. Dev. Biol. 312,
1328 (2007). doi: 10.1016/j.ydbio.2007.08.059;
pmid: 17961534
27. Y. Hinits et al., Defective cranial skeletal development, larval
lethality and haploinsufficiency in Myod mutant zebrafish.
Dev. Biol. 358, 102112 (2011). doi: 10.1016/
j.ydbio.2011.07.015; pmid: 21798255
28. Y. Hinits, D. P. Osborn, S. M. Hughes, Differential requirements
for myogenic regulatory factors distinguish medial and lateral
somitic, cranial and fin muscle fibre populations. Development SCIENCE

Downloaded from on July 8, 2016














136, 403414 (2009). doi: 10.1242/dev.028019;

pmid: 19141670
P. Hasty et al., Muscle deficiency and neonatal death in mice
with a targeted mutation in the myogenin gene. Nature 364,
501506 (1993). doi: 10.1038/364501a0; pmid: 8393145
K. Kikuchi et al., Primary contribution to zebrafish heart
regeneration by gata4+ cardiomyocytes. Nature 464, 601605
(2010). doi: 10.1038/nature08804; pmid: 20336144
G. E. Hollway et al., Whole-somite rotation generates muscle
progenitor cell compartments in the developing zebrafish
embryo. Dev. Cell 12, 207219 (2007). doi: 10.1016/
j.devcel.2007.01.001; pmid: 17276339
F. Stellabotte, B. Dobbs-McAuliffe, D. A. Fernndez, X. Feng,
S. H. Devoto, Dynamic somite cell rearrangements lead to
distinct waves of myotome growth. Development 134,
12531257 (2007). doi: 10.1242/dev.000067; pmid: 17314134
M. T. Webster, C. M. Fan, c-MET regulates myoblast motility
and myocyte fusion during adult skeletal muscle regeneration.
PLOS ONE 8, e81757 (2013). doi: 10.1371/journal.
pone.0081757; pmid: 24260586
L. Haines et al., Met and Hgf signaling controls hypaxial muscle
and lateral line development in the zebrafish. Development 131,
48574869 (2004). doi: 10.1242/dev.01374; pmid: 15342468
T. E. Hall et al., The zebrafish candyfloss mutant implicates
extracellular matrix adhesion failure in laminin a2-deficient
congenital muscular dystrophy. Proc. Natl. Acad. Sci. U.S.A.
104, 70927097 (2007). doi: 10.1073/pnas.0700942104;
pmid: 17438294
H. Zhang, J. E. Anderson, Satellite cell activation and
populations on single muscle-fiber cultures from adult
zebrafish (Danio rerio). J. Exp. Biol. 217, 19101917 (2014).
doi: 10.1242/jeb.102210; pmid: 24577448
N. A. Dumont et al., Dystrophin expression in muscle stem
cells regulates their polarity and asymmetric division.
Nat. Med. 21, 14551463 (2015). doi: 10.1038/nm.3990;
pmid: 26569381
F. Ruf-Zamojski, V. Trivedi, S. E. Fraser, L. A. Trinh, Spatiotemporal differences in dystrophin dynamics at mRNA and
protein levels revealed by a novel FlipTrap line. PLOS ONE 10,
e0128944 (2015). doi: 10.1371/journal.pone.0128944;
pmid: 26083378
Y. A. Pan et al., Zebrabow: Multispectral cell labeling for cell
tracing and lineage analysis in zebrafish. Development 140,
28352846 (2013). doi: 10.1242/dev.094631; pmid: 23757414
P. D. Nguyen et al., Haematopoietic stem cell induction by
somite-derived endothelial cells controlled by meox1. Nature
512, 314318 (2014). doi: 10.1038/nature13678;
pmid: 25119043


41. A. L. Siegel, K. Atchison, K. E. Fisher, G. E. Davis,

D. D. Cornelison, 3D timelapse analysis of muscle satellite cell
motility. Stem Cells 27, 25272538 (2009). doi: 10.1002/
stem.178; pmid: 19609936
42. Y. H. Chen et al., Multiple upstream modules regulate zebrafish
myf5 expression. BMC Dev. Biol. 7, 1 (2007). doi: 10.1186/1471213X-7-1; pmid: 17199897
43. N. J. Cole et al., Development and evolution of the muscles of
the pelvic fin. PLOS Biol. 9, e1001168 (2011). doi: 10.1371/
journal.pbio.1001168; pmid: 21990962
44. J. Berger et al., Loss of Tropomodulin4 in the zebrafish mutant
trge causes cytoplasmic rod formation and muscle weakness
reminiscent of nemaline myopathy. Dis. Model. Mech. 7,
14071415 (2014). doi: 10.1242/dmm.017376; pmid: 25288681
45. S. Elworthy, M. Hargrave, R. Knight, K. Mebus, P. W. Ingham,
Expression of multiple slow myosin heavy chain genes reveals a
diversity of zebrafish slow twitch muscle fibres with differing
requirements for Hedgehog and Prdm1 activity. Development 135,
21152126 (2008). doi: 10.1242/dev.015719; pmid: 18480160
46. H. Pisharath, J. M. Rhee, M. A. Swanson, S. D. Leach,
M. J. Parsons, Targeted ablation of beta cells in the embryonic
zebrafish pancreas using E. coli nitroreductase. Mech. Dev.
124, 218229 (2007). doi: 10.1016/j.mod.2006.11.005;
pmid: 17223324
47. M. Westerfield, The Zebrafish Book. A Guide for the Laboratory
Use of Zebrafish (Danio rerio) (Univ. of Oregen Press, ed. 3, 1995).
48. H. Kikuta, K. Kawakami, Transient and stable transgenesis
using tol2 transposon vectors. Methods Mol. Biol. 546, 6984
(2009). doi: 10.1007/978-1-60327-977-2_5; pmid: 19378098
49. E. Busch-Nentwich et al., Sanger Institute Zebrafish Mutation
Resource targeted knock-out mutants phenotype and image
data submission, ZFIN Direct Data Submission (Sanger
Institute Zebrafish Mutation Resource, MPI Dresden, and
Hubrecht laboratory, 2010).
50. K. M. Kwan et al., The Tol2kit: A multisite gateway-based
construction kit for Tol2 transposon transgenesis constructs.
Dev. Dyn. 236, 30883099 (2007). doi: 10.1002/dvdy.21343;
pmid: 17937395
51. C. Otten, S. Abdelilah-Seyfried, Laser-inflicted injury of
zebrafish embryonic skeletal muscle. J. Vis. Exp. 71, e4351
(2013). pmid: 23407156
52. H. Pisharath, M. J. Parsons, Nitroreductase-mediated cell
ablation in transgenic zebrafish embryos. Methods Mol. Biol.
546, 133143 (2009). doi: 10.1007/978-1-60327-977-2_9;
pmid: 19378102
53. H. C. Seo, B. O. Saetre, B. Hvik, S. Ellingsen, A. Fjose, The
zebrafish Pax3 and Pax7 homologues are highly conserved,
encode multiple isoforms and show dynamic segment-like







expression in the developing brain. Mech. Dev. 70, 4963

(1998). doi: 10.1016/S0925-4773(97)00175-5; pmid: 9510024
J. A. Groves, C. L. Hammond, S. M. Hughes, Fgf8 drives
myogenic progression of a novel lateral fast muscle fibre
population in zebrafish. Development 132, 42114222 (2005).
doi: 10.1242/dev.01958; pmid: 16120642
Y. Hinits, D. P. S. Osborn, J. J. Carvajal, P. W. J. Rigby,
S. M. Hughes, Mrf4 (myf6) is dynamically expressed in
differentiated zebrafish skeletal muscle. Gene Expr. Patterns 7,
738745 (2007). doi: 10.1016/j.modgep.2007.06.003;
pmid: 17638597
E. S. Weinberg et al., Developmental regulation of zebrafish
MyoD in wild-type, no tail and spadetail embryos. Development
122, 271280 (1996). pmid: 8565839
P. Chapouton et al., Notch activity levels control the balance
between quiescence and recruitment of adult neural stem
cells. J. Neurosci. 30, 79617974 (2010). doi: 10.1523/
JNEUROSCI.6170-09.2010; pmid: 20534844
M. Delous et al., Sox9b is a key regulator of pancreaticobiliary
ductal system development. PLOS Genet. 8, e1002754 (2012).
doi: 10.1371/journal.pgen.1002754; pmid: 22719264
J. Berger, T. Sztal, P. D. Currie, Quantification of birefringence
readily measures the level of muscle damage in zebrafish.
Biochem. Biophys. Res. Commun. 423, 785788 (2012).
doi: 10.1016/j.bbrc.2012.06.040; pmid: 22713473


We thank P. Ingham, A. Schier, G. Lieschke, L. Bally-Cuif, and the

Zebrafish International Resource Center for the supply of fish
strains and reagents; R. Cheney for technical advice; N. Cole for
technical assistance; and C. Marcelle and T. Rando for comments
on the manuscript. This work was supported by a National Health
and Medical Research Council of Australia grant to P.D.C. The
Australian Regenerative Medicine Institute is supported by funds
from the state government of Victoria and the Australian federal
government. cmet transgenic lines are available from P.D.C. under
a material transfer agreement with Monash University.

Figs. S1 to S10
Movies S1 to S10
8 December 2015; accepted 10 May 2016
Published online 19 May 2016

8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Asymmetric division of clonal muscle stem cells coordinates

muscle regeneration in vivo
David B. Gurevich, Phong Dang Nguyen, Ashley L. Siegel, Ophelia
V. Ehrlich, Carmen Sonntag, Jennifer M. N. Phan, Silke Berger,
Dhanushika Ratnayake, Lucy Hersey, Joachim Berger, Heather
Verkade, Thomas E. Hall and Peter D. Currie (May 19, 2016)
Science 353 (6295), . [doi: 10.1126/science.aad9969] originally
published online May 19, 2016

Dividing asymmetrically to fix muscle

Resident tissue stem cells called satellite cells repair muscle after injury. However, how satellite
cells operate inside living tissue is unclear. Gurevich et al. exploited the optical clarity of zebrafish
larvae and used a series of genetic approaches to study muscle injury. After injury, satellite cells divide
asymmetrically to generate a progenitor pool for muscle replacement and at the same time ''self-renew''
the satellite stem cell. This results in regeneration that is highly clonal in nature, validating many
decades of in vitro analyses examining the regenerative capacity of skeletal muscle.
Science, this issue p. 136

This copy is for your personal, non-commercial use only.

Article Tools


Visit the online version of this article to access the personalization and
article tools:
Obtain information about reproducing this article:

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week
in December, by the American Association for the Advancement of Science, 1200 New York
Avenue NW, Washington, DC 20005. Copyright 2016 by the American Association for the
Advancement of Science; all rights reserved. The title Science is a registered trademark of AAAS.

Downloaded from on July 8, 2016

Editor's Summary



A metal-organic frameworkbased
splitter for separating propylene
from propane
A. Cadiau,1* K. Adil,1* P. M. Bhatt,1 Y. Belmabkhout,1 M. Eddaoudi1,2
The chemical industry is dependent on the olefin/paraffin separation, which is mainly
accomplished by using energy-intensive processes. We report the use of reticular
chemistry for the fabrication of a chemically stable fluorinated metal-organic
framework (MOF) material (NbOFFIVE-1-Ni, also referred to as KAUST-7). The bridging
of Ni(II)-pyrazine square-grid layers with (NbOF5)2 pillars afforded the construction
of a three-dimensional MOF, enclosing a periodic array of fluoride anions in contracted
square-shaped channels. The judiciously selected bulkier (NbOF5)2 caused the
looked-for hindrance of the previously free-rotating pyrazine moieties, delimiting the
pore system and dictating the pore aperture size and its maximum opening. The
restricted MOF window resulted in the selective molecular exclusion of propane from
propylene at atmospheric pressure, as evidenced through multiple cyclic
mixed-gas adsorption and calorimetric studies.

lefin/paraffin separation is a critical separation for the chemical industry. It is mainly accomplished by means of cryogenic
distillation, which accounts for nearly 3%
[120 teraBritish thermal units (TBtu)/year]
of the total energy used for separations (1). Exploratory olefin/paraffin separation studies that
used conventional adsorbents such as zeolites
and ultramicroporous carbon revealed the lack
of a suitable adsorbent with the appropriate
pore system (functionality and/or aperture size)
that could selectively discriminate and separate
the two relevant vapors with a noticeably reduced energy footprint (2).
Propylene is a prime olefin raw material for
petrochemical production, second in importance
to ethylene, and is an essential building block for
the manufacturing of various chemicals, including polypropylene (3). Propylene purity guides its
end use because polypropylene requires a highpurity propylene with 99.5 weight % (wt %) minimum (polymer-grade specifications), whereas
chemical-grade specifications typically involve a
propylene purity of 92% or more. The purity of
propylene primarily depends on the removal of
propane; this separation process is energy-intensive
and conventionally dominated by the cryogenic

Division of Physical Sciences and Engineering, Advanced

Membranes and Porous Materials Center, Functional Materials
Design, Discovery and Development Research Group (FMD3),
King Abdullah University of Science and Technology (KAUST),
Thuwal 23955-6900, Kingdom of Saudi Arabia. 2Department of
Chemistry, University of South Florida, 4202 East Fowler
Avenue, Tampa, FL 33620, USA.

*These authors contributed equally to this work. Corresponding

author. Email:


distillation because of the close-boiling points

and the slight variance in the condensabilities of
the two components. Researchers have been exploring various separation methods, including (i)
the p complexationdriven separation (as adsorbent or facilitated membrane) (4, 5) or (ii)
separation based on kinetics by using physical
adsorbents (zeolites A) or membranes, driven
mainly by the difference in diffusion rates into
the pore system (6, 7). Thus far, adsorbents and
membranes reported for this task used a partial
size-shape sievingbased separation process and
displayed low to moderate separation factors, as
exemplified in the case of zeolites (8, 9) and carbon molecular sieves (7) encompassing contracted
pore sizes. The full extraction of propylene from
propane by using adsorbent-based separating
agents has yet to be demonstrated and accomplished. This is due to (i) the close structural dimensions of propylene and propane molecules,
in addition to their associated similar physical
properties, and (ii) difficulties in fine-tuning the
pore aperture size in 0.2 to 1 scale increments
in a given ultramicropore adsorbent material
with a suitable pore system size ranging between
3 and 5 to 6 .
Ideally, porous materials constructed by means
of the molecular building block approach offer
potential to attain the requisite separation through
selective molecular exclusion, in which the molecular building blocks could be adopted in shape,
size, and functionality so as to produce a suitable
pore aperture size that can selectively separate
molecules with almost identical physical properties. We recently succeeded in selectively separating branched paraffins from normal paraffins by

fine-tuning a metal-organic framework (MOF)

aperture size to completely exclude branched
paraffins (10).
MOF materials represent a tunable class of
hybrid solid-state materials that consist of metal
nodes or metal clusters coordinated to multifunctional organic linkers (1113). MOFs have received
considerable attention as adsorbents for gas storage applications (1417) and have demonstrated
high potential for equilibrium- and/or kinetic-based
gas separations, such as CO2/N2 or CO2/H2 (1822).
MOF chemistry has permitted the prospective
fabrication of porous materials with a high degree
of structural predictability and subsequently a
precise control of window apertures size and
shape at the molecular level, in contrast to those
of zeolites, carbon, and polymeric-based materials (10).
We demonstrate a full molecular exclusion of
propane from propylene at standard ambient
temperature and pressure using a fluorinated
porous MOF adsorbent, KAUST-7. The reticular
chemistry (23) approach was used to construct
this material based on pillared square-grids
strategytwo-dimensional nets based on linked
metal nodes that are pillared via a specific inorganic molecular building block in the third dimension so as to form a three-periodic net with a
primitive cubic topology (Fig. 1).
Previously, we applied coordination chemistry
principles to demonstrate that the introduction
of specific cationsZn2+, Ni2+, and Cu2+into
the metal-pyrazine square-grid layers has an impact on the unit cell volume and more precisely
on the pore size, which resulted in an enhancement of the CO2-framework energy interaction.
This family of materials possesses distinct CO2
capture properties, especially at very low concentrations, which is ideal for postcombustion capture and air-capture application (24). Nevertheless,
the SIFSIX-3-Ni (Fig. 1, C and E) was found to
adsorb both propylene and propane.
Substitution of the metal node in the SIFSIX
(24) platform permitted the fine-tuning of the
aperture size and pore contraction and subsequent CO2 capture properties. In order to explore
the possibility of further tuning this platform, we
closely examined the structural features of the
SIFSIX platform (Figs. 1 and 2) (24) and identified an alternative pathway to potentially further fine-tune the pore aperture size. The hindered
free rotation of the pyrazine-bridging ligands, by
altering the nature, shape, and dimensions of the
pillars used, dictates the pore aperture size and its
maximum opening. This approach offers the potential to select for the passing or blocking of
specific probe molecules.
To construct the targeted material with these
envisioned structural features, we opted to maintain the Ni(II) as a six-connected node, connecting the pyrazine ligands in square gridlike layers,
and to modify the inorganic pillars. We anticipated
that using a pillar with a bigger cation (Nb5+ instead of Si4+) would result in a relatively shorter
proximal distance between the adjacent fluorine
centers and consequently contract the aperture
size of the channels (Fig. 2). Reactions between
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




8 JULY 2016 VOL 353 ISSUE 6295

at 77 K, indicating the restricted access to N2

at this low cryogenic temperature due to the
contracted pore aperture size. On the other hand,

adsorption studies performed at room temperature by using CO2 as the adsorbate molecule revealed that NbOFFIVE-1-Ni is microporous,

Fig. 1. Structure description of NbOFFIVE-1-Ni highlighting the building blocks arrangement

and its comparison with the parent SIFSIX-3-Ni. (A) Illustration of the square-shaped arrangement
in the Ni-pyrazine (4,4) square grid that is further pillared by inorganic blocks [(NbOF5)2 or (SiF6)2]
to generate a three-dimensional MOF with a primitive cubic topology. (B) Crystal structure of NbOFFIVE-1-Ni
at 100 K, showing the titling of pyrazine molecules. (C) Crystal structure of SIFSIX-3-Ni at 298 K.
(D) Simulation of the maximum open structure of NbOFFIVE-1-Ni, showing a theoretical window size
of 4.752(1) . (E) Simulation of the maximum open structure of SIFSIX-3-Ni, showing a theoretical
window size of 7.314(1) .

Fig. 2. Structure comparison between SIFSIX-3-Ni and NbOFFIVE-1-Ni showing the impact of
the use of a relatively bigger inorganic pillar [(NbOF5)2 instead of (SiF6)2]. (A) [010]-projection
view of SIFSIX-3-Ni, showing the perfectly aligned pyrazine molecules. (B) [010]-projection view of
NbOFFIVE-1-Ni, showing the titling of pyrazine molecules due to the use of the bigger (NbOF5)2
building block. The shorter distance between two (NbOF5)2 building blocks from the two neighboring layers [4.000(1) versus 4.417(1) in the case of (SiF6)2 pillars in the SIFSIX-3-Ni] induces a
titling of pyrazine molecules and consequently a decrease in the pore aperture size, comparatively
with that of SIFSIX-3-Ni. SCIENCE

Downloaded from on July 8, 2016

Ni(NO3)26H2O, Nb2O5, HFaq, and pyrazine (pyr)

in water yielded purple square-shaped crystals.
The single-crystal x-ray diffraction data revealed the
anticipated primitive-cubic structure, formulated as
NiNbOF5(pyr)22H2O, in good agreement with the
elemental analysis and the thermal gravimetric
analysis (TGA) results (supplementary materials,
materials and methods, and fig. S6). The phase
purity of resultant fluorinated MOF material, referred to as NbOFFIVE-1-Ni or KAUST-7, was
confirmed by using a full pattern matching by
means of the Le Bail method (fig. S1).
The assignment of fluorine atoms in equatorial
positions within the pillar has been previously
demonstrated in similar materials (25, 26). The
particular use of (NbOF5)2 as the pillaring inorganic building block, instead of (SiF6)2, resulted in a longer metal-fluorine distance (1.95 )
and a subsequent tilting of pyrazine molecules
(Figs. 1 and 2) (19, 24). Analysis of the NbOFFIVE1-Ni structure (collected at 100 K) revealed the
plausible smallest pore window opening associated with the relatively hindered rotation of
the (NbOF5)2 pillars, reinforced by the presence
of hydrogen bond interactions (Fig. 2 and fig. S3).
As a result, the hydrogen atoms of the pyrazine
linkers circumference the resultant square-shaped
channels with a reduced pore aperture size of
3.0471(1) [versus 5.032(1) for SIFSIX-3-Ni]
(Fig. 1B), prohibiting the diffusion of any molecule other than water (numbers in parentheses
indicate the 1 SD uncertainty in the final digits).
In order to gain a better insight on the plausible
rotation and tilting of the pyrazine linker, and
subsequently derive a relative maximum opening of the window, providing a gate limit for the
largest molecule to pass through, we collected and
analyzed the same structure at room temperature.
No structural differences were detected from singlecrystal x-ray data collected at room temperature
and the cryogenic temperature. The thermal ellipsoids analysis revealed a certain degree of rotation of (NbOF5)2 pillars, whereas the pyrazine
molecules were crytallographically well localized.
The (NbOF5)2 pillars and pyrazine molecules
were interconnected through a network of strong
hydrogen bonding interactions [FH 2.483(1) ].
The noticeably small pore aperture size, derived
from the crystal structure at room temperature,
suggests that the passing of the propylene molecules is governed by the plausible gate opening
associated with the extra tilting of the pyrazine
molecules under propylene gas stimuli. The pore
aperture size is regulated by the presence or
omission of the steric hindrance between pyrazine molecules and (NbOF5)2 pillars, and the
lack of a precise kinetic diameter for propylene
in the open literature makes precise quantification difficult. Presumably, the plausible complete
omission of the occurring steric hindrance can
afford a theoretical maximum pore aperture size
of 4.752(1) (Fig. 1D).
To further confirm the restricted pore size opening owing to the hindered rotation of pyrazine
ligands at low temperature, we performed adsorption studies on the fully evacuated NbOFFIVE-1-Ni.
As anticipated, NbOFFIVE-1-Ni did not adsorb N2

with an apparent surface area of 280 m2/g

and an estimated pore volume of 0.095 cm3/g
(fig. S8).
The permanent porosity and the associated
restricted pore aperture size [maximum 4.752(1) ]
prompted us to assess and evaluate the adsorption properties of NbOFFIVE-1-Ni for various
key probe molecules characterized by different
shapes and a comparatively larger kinetic diameter. Specifically, adsorption properties of molecules that are relevant to both industrial gases
upgrading and the challenging olefin/paraffin
separationnamely methane (CH4), ethane (C2H6),
ethylene (C2H4), propylene (C3H6), and propane
(C3H8)were evaluated at room temperature
and atmospheric pressures.
Before assessing the NbOFFIVE-1-Ni as a potential separating agent, we evaluated its stability
against water vapor and H2S by monitoring the
powder x-ray diffraction (PXRD) patterns upon
exposure to various humidity levels and H2S concentrations (Fig. 3A and fig. S4). Evidently, no loss
of crystallinity and no phase change were observed during the exposure of the material to
humidity from 5 to 95% (fig. S4), H2O (fig. S7),
or H2S (up to 10%) (fig. S9). The hydrolytic sta-

bility of KAUST-7 is supported by the retention

of the material CO2 adsorption uptake and structural integrity after its immersion in aqueous
solution for more than 6 months. The performance of the material was not altered as the maximum CO2 uptake was reached (fig. S10). Moreover,
the NbOFFIVE-1-Ni is proven to be highly thermal
stable, as evidenced through thermogravimetric
analysis (TGA) and variable temperature PXRD
patterns (fig. S5 and S6).
Single-gas adsorption data revealed that the
NbOFFIVE-1-Ni channels with restricted aperture size allowed the adsorption of C3H6 but did
not permit the C3H8 to diffuse/adsorb into the
pore system at 298 K up to ~1 bar (Fig. 3B). Considerably, C3H6/C3H8 50/50 mixed-gas adsorption data, collected at 298 K (up to 0.5 bar partial
pressure of C3H6), overlaid with the pure C3H6
adsorption isotherm (Fig. 3B), which supports
the molecular exclusion of propane from propylene. The concomitant pore aperture size and
shape expressed in this MOF adsorbent provide
the requisite size and shape cut-off in adsorption,
resulting in the observed C3H6/C3H8 selectivity.
The selectivity was further confirmed by performing C3H6/C3H8 50/50 mixed-gas column break-

Fig. 3. Evaluation of the chemical stability and propylene/propane separation ability of NbOFFIVE1-Ni. (A) Powder x-ray diffraction experiments showing the stability of NbOFFIVE-1-Ni after exposure to
water and H2S. (B) The pure C3H8 (pink), pure C3H6 (purple), and equimolar mixture of C3H6/C3H8
50/50 (orange) isotherms of NbOFFIVE-1-Ni have been collected at 298 K, demonstrating the full propylene
from propane sieving ability of this adsorbent at 1 bar. (C) C3H6/C3H8 50/50 mixed-gas experiment using a
packed column bed at 298 K and a 1 bar total pressure and 4 cm3/min total flow, confirming the infinite
C3H6/C3H8 separation factor. (D) Calorimetric measurements of C3H8 and C3H6 adsorption on NbOFFIVE1-Ni were performed so as to quantify the heat of adsorption of propylene and to reaffirm the exclusion/no
adsorption of propane.


through experiments (Fig. 3C), imitating the

real conditions for the C3H6/C3H8 separation
process, at room temperature and 1 bar in a
packed column bed of ~1.4 g of NbOFFIVE-1-Ni.
By using 4 cm3/min total gas flow, C3H8 was not
adsorbed in the packed column bed, whereas
pure C3H6 was retained for ~480 s (Fig. 3C). Additionally, mixed-gas column breakthrough experiments were performed in dilute conditions
by using N2 as a carrier inert gas, namely C3H6/
C3H8/N2 in 5/5/90 (fig. S25) and C3H6/C3H8/N2
in 25/25/90 (fig. S26) mixtures. The pure C3H6
was retained in the packed column bed, whereas
N2 and C3H8 were not adsorbed or retained by
NbOFFIVE-1-Ni. The regeneration and activation of the saturated adsorbent, desorption over
a 10-min period, showed solely the propylene signal and thus confirmed the nonadsorption and
nonretention of the propane in the bed (fig. S27).
To support and confirm the complete molecular exclusion of C3H8 and the sole adsorption of
C3H6, simultaneous calorimetric and gravimetric
measurements (thermogravimetrydifferential
scanning calorimetry) were performed at 1 bar.
This confirmed the complete exclusion of propane from propylene, as evidenced by the lack
of a detectable heat of adsorption in the case of
C3H8, as compared with the heat of adsorption
for C3H6 of 57.4 kJ/mol (Fig. 3D). High-pressure
adsorption studies confirmed the nondetectable adsorption of C3H8 below 1.5 bar, and only
minor propane uptake was observed at ~1.5 bar
(~0.1 mol/kg) (fig. S11).
Prior C3H6/C3H8 adsorption studiesusing zeolite,
carbon molecular sieves (CMSs), or MOFsrevealed
the plausible equilibrium- and/or kinetic-based
separation but with low to moderate separation
factors (7, 2729). Correspondingly, the evaluation of zeolite molecular sieves for the propylene/
propane separation under similar experimental
conditions as the NbOFFIVE-1-Ni adsorbent revealed (i) the very poor C3H6/C3H8 separation
(fig. S12) for zeolite 4A at 1 bar total pressure, as a
result of the very slow C3H6 adsorption kinetics
(fig. S13), and (ii) the extremely low selectivity
of nearly 2 for zeolite 5A, despite its associated fast
adsorption kinetics relative to that of NbOFFIVE1-Ni (fig. S13).
The deployment of NbOFFIVE-1-Ni as a splitter agent or adsorbent, permitting the complete
sieving of C3H8 from C3H6, offers (i) a simplified
separation process based on a concentration swing
recycling mode (CSRM) or a vacuum swing recycling mode (VSRM), in which the ideal working
C3H6 capacity can be accomplished by performing
a desorption step with an inert gas (such as He or
N2) purge at 1.2 bar or by simply reducing the
pressure from 1.2 bar to 0.01 bar (Fig. 3B); and (ii)
the ability to eliminate the energy-demanding
high-pressure steps used in the case of the zeolite
4A adsorbentpressurization (step 2), purge (with
N2, step 3), and cocurrent blow down (step 4) will
not be required in the projected concentration
swing adsorption (CSA) or vacuum swing adsorption (VSA) system when using the NbOFFIVE-1-Ni
adsorbent (Fig. 4B). The implementation of
the VSA system based on NbOFFIVE-1-Ni as an
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



adsorbent offers the potential to considerably

reduce the energy penalty associated with the
conventional C3H6/C3H8 separation, and valuably
recover both C3H6 and C3H8 separately in a highpurity grade.
Correspondingly, we performed subsequent
mixed-gas (C3H6/C3H8 at 50/50) column breakthrough measurements in order to corroborate
the preservation of the adsorption properties
and separation performance of the NbOFFIVE1-Nithe propylene adsorption uptake and the
full molecular exclusion of propane from propylene at standard ambient temperature and pressure.
The multiple adsorption/desorption measurements
(over 10 cycles) by use of CSRM revealed that
NbOFFIVE-1-Ni maintained its propylene adsorption capacity and its full molecular exclusion of
propane (Fig. 3C and figs. S14 to S23). Conversely,
the deployment of the aforementioned CSRM
protocol under the same operating conditions in
the case of the evaluated zeolite molecular sieves
revealed a noticeable decrease in the C3H6 retention time by the zeolite 5A, which is equivalent to
a 300% reduction after only four adsorption/
desorption cycles (fig. S24). In the case of zeolite 4A, the low propylene uptake associated
with its very slow diffusion limited the scope of
zeolite 4A for the C3H6/C3H8 separation under
atmospheric conditions.
Detailed analysis of the data, for the C3H6/C3H8
50/50 mixed-gas adsorption cycles in a bed composed of 1.4 g of NbOFFIVE-1-Ni, indicated a C3H6
uptake of ~0.6 mol/kg for a given cycle based on
an 8-min adsorption followed by a 10-min desorption (Fig. 3C) when using CSRM. This result
pinpoints the appropriateness of the NbOFFIVE-1Ni as a stable separating agent for propylene/
propane, with a pronounced propylene uptake/
recovery of ~2 mol/kg/hour. The NbOFFIVE-1-Ni
adsorbent offers potential to effectively separate
propylene from propane with a reduced energy
footprint by using a CSA. The zeolite 4A, when
using VSA at high pressure and 423 K, offers only
a limited 26% recovery for a propylene capacity of
1.03 mol/kg/hour (0.13 mol/kg per cycle), with a
97% purity (30).
We have demonstrated the successful use of
reticular chemistry to deliberately fabricate a
fluorinated MOF adsorbent, with controlled pore

8 JULY 2016 VOL 353 ISSUE 6295

aperture size and shape, for the selective molecular exclusion of propane from propylene at
standard ambient temperature and pressure, up
to 1.5 bar. This complete molecular sieving of
propane from propylene was supported by a
combined comprehensive study on the basis of
single-gas, mixed-gas adsorption and calorimetricgravimetric experiments. The NbOFFIVE-1-Ni adsorbent possesses a noticeably high chemical
stabilityas demonstrated by its tolerance to
water vapor and hydrogen sulfideand high
thermal stability. The full molecular exclusion of
propane by NbOFFIVE-1-Ni adsorbent provides
the potential to deploy a simplified and energyefficient concentration swing adsorption system
that does not require energy-demanding steps
such as pressurization, high-pressure purge, or
cocurrent blow down.
This study illustrates the merit of the molecular building block approach in MOF chemistry
for the ultrafine-tuning of the pore aperture size
and shape to address important gas separations.
Evidently, high porosity is not a sorbent perquisite for an effective and efficient gas separation;
a nonporous material to the N2 adsorbate at
cryogenic temperaturea routine probe to assess
the microporosity of a given adsorbent such as
KAUST-7can possess distinctive gas separation
properties that can be revealed and disclosed
through the close examination of the material
structural features and the subsequent deeper
examination and evaluation of its gas adsorption properties at room temperature.

1. U.S. Department of Energy (DOE), Materials for Separation

Technology: Energy and Emission Reduction Opportunities
(DOE, 2005).
2. P. Bai et al., Nat. Commun. 6, 5912 (2015).
3. A. Chauvel, G. Lefebvre, in Petrochemical Processes
(Gulf Publishing, 1989), pp. 199208.
4. D. J. Safarik, R. B. Eldridge, Ind. Eng. Chem. Res. 37,
25712581 (1998).
5. R. T. Yang, E. S. Kikkinides, AIChE J. 41, 509517
6. M. C. Campo et al., Separ. Purif. Tech. 103, 6070
7. J. Liu et al., Carbon 85, 201211 (2015).
8. C. A. Grande, J. Gascon, F. Kapteijn, A. E. Rodrigues, Chem.
Eng. J. 160, 207214 (2010).
9. C. A. Grande, A. E. Rodrigues, Ind. Eng. Chem. Res. 44,
88158829 (2005).

10. A. H. Assen et al., Angew. Chem. Int. Ed. Engl. 54, 1435314358
11. G. Frey, Chem. Soc. Rev. 37, 191214 (2008).
12. S. Kitagawa, R. Kitaura, S. Noro, Angew. Chem. Int. Ed. Engl.
43, 23342375 (2004).
13. M. Eddaoudi et al., Acc. Chem. Res. 34, 319330
14. A. R. Millward, O. M. Yaghi, J. Am. Chem. Soc. 127,
1799817999 (2005).
15. D. Fairen-Jimenez et al., Chem. Commun. 48, 1049610498
16. Y. Peng et al., J. Am. Chem. Soc. 135, 1188711894
17. Q. Yang et al., Chem. Commun. 48, 98319833 (2012).
18. D.-X. Xue et al., J. Am. Chem. Soc. 137, 50345040
19. P. Nugent et al., Nature 495, 8084 (2013).
20. K. Sumida et al., Chem. Rev. 112, 724781 (2012).
21. S. Xiang et al., Nat. Commun. 3, 954 (2012).
22. Z. R. Herm, R. Krishna, J. R. Long, Micropor. Mesopor. Mater.
151, 481487 (2012).
23. O. M. Yaghi et al., Nature 423, 705714 (2003).
24. O. Shekhah et al., Nat. Commun. 5, 4228 (2014).
25. H. K. Izumi, J. E. Kirsch, C. L. Stern, K. R. Poeppelmeier, Inorg.
Chem. 44, 884895 (2005).
26. R. Gautier, M. D. Donakowski, K. R. Poeppelmeier, J. Solid State
Chem. 195, 132139 (2012).
27. O. Shekhah et al., Chem. Commun. 50, 20892092
28. E. D. Bloch et al., Science 335, 16061610 (2012).
29. X. Yang, B. H. Toby, M. A. Camblor, Y. Lee, D. H. Olson, J. Phys.
Chem. B 109, 78947899 (2005).
30. F. A. Da Silva, A. E. Rodrigues, Ind. Eng. Chem. Res. 40,
57585774 (2001).

Research reported in this publication was solely performed

in KAUST and exclusively supported by KAUST funds and the
KAUST center collaborative funding grants CCF/1/1972-02-01
and CCF/1/1972-8-01. The data reported in the paper are
presented in the supplementary materials. Crystal structures of
NbOFFIVE-1-Ni collected at 100 K and room temperature are
available free of charge from the Cambridge Crystallographic
Data Centre under reference nos. CCDC 1477136 and 1447953.
A.C., K.A., Y.B., M.E., and KAUST have filed provisional patent
applications (4053-023PCT1 and 4053-024PCT1) that relate
to highly stable fluorinated MOF absorbents as tunable
platforms for gas/vapor separation.
Materials and Methods
Supplementary Text
Figs. S1 to S29
Tables S1 and S2
References (3135)
CIF Files (Crystal Structure)
7 March 2016; accepted 25 May 2016
10.1126/science.aaf6323 SCIENCE

Downloaded from on July 8, 2016

Fig. 4. Separation process based on concentration swing recycling mode and a simplified
CSA or VSA system for the NbOFFIVE-1-Ni molecular sieve. (A) C3H6/C3H8 50/50 mixed-gas
11 cyclic experiments for NbOFFIVE-1-Ni at 298 K
and 1 bar, with 4 cm3/min total flow and using a
CSRM system, with He as a purge gas for cycles
2 to 10, whereas the sample was activated for
8 hours at 105C under vacuum for cycle 1 and
cycle 11. (B) Principle of using NbOFFIVE-1-Ni as
a C3H8 sieve in two-step CSA or VSA (red line,
steps 1 and 5) as compared with a pressure swing
adsorption (PSA) system using the zeolite 4A
(steps 1 through 5).


Pore chemistry and size control in

hybrid porous materials for acetylene
capture from ethylene
Xili Cui,1* Kaijie Chen,2* Huabin Xing,1 Qiwei Yang,1 Rajamani Krishna,3
Zongbi Bao,1 Hui Wu,4 Wei Zhou,4 Xinglong Dong,5 Yu Han,5 Bin Li,6 Qilong Ren,1
Michael J. Zaworotko,2 Banglin Chen6
The trade-off between physical adsorption capacity and selectivity of porous materials
is a major barrier for efficient gas separation and purification through physisorption.
We report control over pore chemistry and size in metal coordination networks with
hexafluorosilicate and organic linkers for the purpose of preferential binding and orderly
assembly of acetylene molecules through cooperative host-guest and/or guest-guest
interactions. The specific binding sites for acetylene are validated by modeling and
neutron powder diffraction studies. The energies associated with these binding
interactions afford high adsorption capacity (2.1 millimoles per gram at 0.025 bar) and
selectivity (39.7 to 44.8) for acetylene at ambient conditions. Their efficiency for the
separation of acetylene/ethylene mixtures is demonstrated by experimental
breakthrough curves (0.73 millimoles per gram from a 1/99 mixture).

n urgent demand for efficient solutions to

challenges in gas separation, sensing, and
storage (17) has spurred research on customdesigned porous materials, termed metalorganic frameworks (MOFs) and/or porous
coordination polymers (PCPs) (8), in which open
lattices are formed from inorganic centers (nodes)
and organic linking groups. These materials can be
designed from first principles and, thanks to their
inherent diversity, afford precise control over pore
chemistry and pore size.
Ideal porous materials for gas separation should
exhibit high selectivity and optimal adsorption capacity for the target gas molecules at relevant conditions. However, the design of new materials that
improve upon existing benchmarks poses a daunting challenge to materials scientists (912). For
example, porous materials are needed for acetylene
(C2H2) capture and separation from ethylene (C2H4)
(1317), industrial processes that are relevant for
the production of polymer-grade C2H2 and C2H4
(the most produced organic compound in the world,
at over 140 million metric tons in 2014). The

Key Laboratory of Biomass Chemical Engineering of

Ministry of Education, College of Chemical and Biological
Engineering, Zhejiang University, Hangzhou 310027, China.
Department of Chemical and Environmental Sciences,
University of Limerick, Limerick, Republic of Ireland.
Van 't Hoff Institute for Molecular Sciences, University
of Amsterdam, Science Park 904, 1098 XH Amsterdam,
Netherlands. 4 Center for Neutron Research, National
Institute of Standards and Technology, Gaithersburg,
MD 20899-6102, USA. 5Advanced Membranes and Porous
Materials Center, Physical Sciences and Engineering
Division, King Abdullah University of Science and
Technology, Thuwal 23955-6900, Saudi Arabia.
Department of Chemistry, University of TexasSan
Antonio, One UTSA Circle, San Antonio, TX 78249-0698,
*These authors contributed equally to this work. Corresponding
author. Email: (H.X.);
(M.J.Z.); (B.C.)


MOF-74 family of compounds has a high density of open metal sites that drive high uptake of
C2H2 but displays low separation selectivities (18).
The MMOF (mixed metalorganic framework)
family has ultramicropores that enable sieving
effects and high separation selectivities but
relatively low uptake of C2H2 (19).
We report that metal coordination networks
with preformed inorganic and organic linkers
SIFSIX-2-Cu-i [SIFSIX, hexafluorosilicate (SiF62); 2,
4,4-dipyridylacetylene; i, interpenetrated)] (20)
and SIFSIX-1-Cu (1, 4,4-bipyridine) (21)can
exhibit exceptional C2H2 capture performance because the geometric disposition of SiF62 moieties
enables preferential binding of C2H2 molecules.
Both materials have pore spaces that enable extremely high C2H2 capture under low pressures,
and they unexpectedly represent new benchmarks
for the highly efficient removal of minor amounts
of C2H2 from C2H4 gas (SIFSIX-2-Cu-i) and mass
separation of C2H2/C2H4 mixtures under ambient
conditions (SIFSIX-1-Cu). We attribute this unprecedented performance to the existence of sweet
spots in pore chemistry and pore size that enable
highly specific recognition and high uptake of
C2H2 to occur in the same material.
In these SIFSIX materials, two-dimensional (2D)
nets of organic ligand and metal node are pillared
with SiF6 anions in the third dimension to form
3D coordination networks that have primitive
cubic topology and, importantly, pore walls lined
by inorganic anions (2023). The pore sizes within
this family of materials can be systematically tuned
by changing the length of the organic linkers, the
metal node, and/or the framework interpenetration.
SIFSIX-1-Cu, SIFSIX-2-Cu, SIFSIX-2-Cu-i, SIFSIX3-Cu (3, pyrazine), SIFSIX-3-Zn, and SIFSIX-3-Ni
have already been studied for their exceptional
CO2 capture performance, but here we report a
study of their C2H2 and C2H4 adsorption from 283

to 303 K. Figure 1, A and B, and figs. S2 to S7 show

dramatically different adsorption behaviors for
C2H2 than those observed for CO2 (figs. S8 and
S9) (24). SIFSIX-2-Cu-i rapidly adsorbs C2H2 at
very low pressure (0.05 bar). Its C2H2 uptake
reaches 2.1 mmol/g at 298 K and 0.025 bar (Fig. 1B),
compared with an uptake of 1.78 mmol/g by
UTSA-100a (17). This performance at low pressure indicates that SIFSIX-2-Cu-i has promise
for capturing C2H2 when it is a minor component
in a gas mixture. SIFSIX-1-Cu exhibits exceptionally high C2H2 uptake (8.5 mmol/g) at 298 K and
1.0 bar. This is not only the highest uptake of
any of the SIFSIX materials, but is also even
higher than that of the previous benchmark,
FeMOF-74 (table S1) (18, 24). As detailed herein,
we attribute the unprecedented performance of
these materials to their hybrid pore chemistry
and optimal pore sizes for binding C2H2.
To understand the C2H2 adsorption isotherms
in these materials, we conducted detailed modeling
studies using first-principles DFT-D (dispersioncorrected density functional theory) calculations.
In SIFSIX-1-Cu, C2H2 molecules are bound through
strong CHF hydrogen (H) bonding (2.017 )
and van der Waals (vdW) interactions with the
4,4-bipyridine linkers (Fig. 1C and fig. S10) (24).
The DFT-Dcalculated static adsorption energy
(DE) is 44.6 kJ/mol. Each unit cell of SIFSIX-1-Cu
contains four equivalent exposed F atoms, and
each exposed F atom binds one C2H2 molecule.
The distance between neighboring adsorbed C2H2
molecules is ideal for them to synergistically interact with each other through multiple Hd+C d
dipole-dipole interactions (Fig. 1C), further enhancing the energy of adsorption. Because four
C2H2 molecules are adsorbed per unit cell, the
DE of C2H2 increases to 47.0 kJ/mol. The strong
binding of C2H2 at F atoms and the geometric
arrangement of SiF62 anions enable the efficient
packing of four C2H2 molecules per unit cell and
very high C2H2 uptake at 298 K and 1.0 bar
(about 4.4 C2H2 molecules per unit cell).
C2H2 adsorption is weaker in the wider-pore
material SIFSIX-2-Cu (10.5 10.5 cavity) than
in SIFSIX-1-Cu (DE, 34.6 versus 44.6 kJ/mol; uptake, 5.3 versus 8.5 mmol/g). The CHF H-bonding
interaction from SiF62 sites is of the same nature
in these isoreticular networks (Fig. 1, C and D).
However, the vdW interaction between C2H2 and
the organic linker in SIFSIX-2-Cu is weak compared
with that in SIFSIX-1-Cu. We attribute this difference to the formers larger pore size and weaker
vdW potential overlap (figs. S10 and S11) (24). Moreover, at high gas uptake, the C2H2 molecules adsorbed on adjacent F sites are too far separated
to have synergistic guest-guest interactions. However, in the twofold interpenetrated structure of
SIFSIX-2-Cu-i, one C2H2 molecule can be simultaneously bound by two F atoms from different
nets through cooperative CHF H-bonding
(2.013 and 2.015 ; Fig. 1E), which enables the
strongest energy of C2H2 binding yet observed in
SIFSIX materials (DE, 52.9 kJ/mol). The strong
adsorption energy of SIFSIX-2-Cu-i contributes
to its extremely high uptake capacity at low pressure. In SIFSIX-3-Zn and SIFSIX-3-Ni, which are
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



Fig. 1. C2H2 and C2H4 adsorption isotherms of the MOFs, DFT-Dsimulated

optimized C2H2 adsorption sites of the MOFs, and neutron crystal structure
of SIFSIX-1-Cu4C2D2. (A and B) Adsorption isotherms of C2H2 (filled circles)
and C2H4 (triangles) in SIFSIX-1-Cu (red), SIFSIX-2-Cu (green), SIFSIX-2-Cu-i
(blue), SIFSIX-3-Zn (light blue), and SIFSIX-3-Ni (orange) at 298 K in two
pressure regions, 0 to 1.0 bar (A) and 0 to 0.05 bar (B). Open circles in (A)

8 JULY 2016 VOL 353 ISSUE 6295

are desorption isotherms of C2H2 (C to F) DFT-Dcalculated C2H2 adsorption

binding sites in SIFSIX-1-Cu (C), SIFSIX-2-Cu (D), SIFSIX-2-Cu-i (E) (the
different nets are highlighted in magenta and green for clarity), and SIFSIX-3Zn (F). Color code: F, red; Si, light blue; C, gray; H, light gray; N, sky blue; Cu, dark
teal; Zn, violet; C (in C2H2 or C2D2), orange. (G and H) Neutron crystal structure of
SIFSIX-1-Cu4C2D2 at 200 K, determined from Rietveld analysis.

Fig. 2. IAST calculations for MOF performance with C2H2/C2H4 mixtures. (A and B) Comparison
of the IAST selectivities of SIFSIX materials versus those of previously reported best-performing materials for
C2H2/C2H4 mixtures. Results for varying C2H2 molar fractions at 100 kPa are shown in (A), and results at
varying pressures for a 1% C2H2 mixture are shown in (B). (C and D) MOF capacity to uptake C2H2 from C2H2/
C2H4 mixtures. Results for varying C2H2 molar fractions at 100 kPa are shown in (C), and results at varying
pressures for a 1% C2H2 mixture are shown in (D).


the strongest CO2 adsorbents, the pore size is

smallest, and C2H2 molecules are primarily adsorbed at a different site in the 1D channel along
the c axis (DE, 50.3 kJ/mol; Fig. 1F). The secondary
adsorption site in SIFSIX-3-Zn (fig. S12) (24) exhibits a much smaller adsorption energy (25.9 kJ/mol),
which results in lower C2H2 uptake compared with
that of SIFSIX-2-Cu-i.
The weakly basic nature of the SiF62sites (acid
dissociation constant pKa, 1.92) and their geometric
disposition enable strong binding with weakly
acidic C2H2 molecules. Because C2H2 is more acidic
than C2H4 (pKa, 25 versus 44) (17), and the geometry of the SIFSIX materials is more optimal for
C2H2 binding, there are much stronger interactions
with C2H2 than with C2H4 (DE in SIFSIX-1-Cu, 44.6
versus 27.2 kJ/mol; DE in SIFSIX-2-Cu-i, 52.9 versus
39.8 kJ/mol). The calculated H-bond distances
between C2H4 and SiF62 sites are 2.541 and 2.186
in SIFSIX-1-Cu and SIFSIX-2-Cu-i, respectively,
which are longer than those between C2H2 and
SiF62 sites (figs. S13 and S14) (24).
To establish the structure of the C2H2 binding
sites through Rietveld structural refinements,
high-resolution neutron powder diffraction data
were collected on C2D2-loaded samples of SIFSIX1-Cu4C2D2 and SIFSIX-2-Cu-i1.7C2D2 at 200 K (figs.
S15 and S16) (24). Each unit cell of SIFSIX-1-Cu is
filled with four C2D2 molecules that are arranged in an ordered planar structure (Fig. 1, G
and H), consistent with the DFT-D modeling
results. CDF H-bonding occurs between C2D2
and SiF6 anions (2.063 ), and Dd+Cd distances between neighboring C2D2 molecules are
3.063 and 3.128 . In SIFSIX-2-Cu-i, each C2H2 SCIENCE

Downloaded from on July 8, 2016

Fig. 3. Simulated and experimental column breakthrough results. (A and B) Simulated column breakthrough curves for C2H2/C2H4 separations with SIFSIX
materials and previously reported best-performing materials [(A), 1/99 mixture; (B), 50/50 mixture]. (C and F) Plots of the amount of C2H2 captured as a
function of tbreak in the simulated column breakthrough [(C), 1/99 mixture; (F), 50/50 mixture]. (D and E) Experimental column breakthrough curves for
C2H2/C2H4 separations with SIFSIX-1-Cu, SIFSIX-2-Cu, and SIFSIX-3-Zn at 298 K and 1.01 bar [(D), 1/99 mixture; (E), 50/50 mixture]. In (A), (B), (D), and
(E), open circles are for C2H4, and filled circles are for C2H2. CA/C0, outlet concentration/feed concentration.

interacts with two SiF6 anions via dual CDF

H-bonding (2.134 ; fig. S17) (24).
The separation of C2H2 from C2H4 is necessary
for the production of high-purity C2H4 and C2H2.
In the production of polymer-grade C2H4, removal
of trace C2H2 (about 1%) from C2H4 gas must meet
the requirement of <40 parts per million (ppm)
C2H2 in the downstream polymerization reaction
(17). Similarly, in the production of polymer-grade
C2H2 by pyrolysis of coal and biomass, the capture
of C2H2 from C2H2/C2H4 mixtures (90/10 to 50/
50, v/v) is a crucial step. Existing methods, such as
solvent absorption and partial hydrogenation of
C2H2 (25), are energy intensive, so there is an urgent need to develop efficient porous materials for
C2H2 capture from C2H4.
To address gas mixture separations, we first
determined the C2H2/C2H4 separation selectivities
of the SIFSIX materials by means of ideal adsorbed
solution theory (IAST) calculations (Fig. 2 and
fig. S20) (24, 26). SIFSIX-2-Cu-i exhibits record
C2H2/C2H4 selectivities (39.7 to 44.8; Fig. 2A), even
greater than those of MMOF-3a (table S1). Because SIFSIX-2-Cu-i adsorbs a high amount of
C2H2 under very low pressures, it not only has
the highest C2H2/C2H4 selectivities (Fig. 2B) but
also the highest C2H2 uptake (Fig. 2D) for C2H2/
C2H4 (1/99) mixtures. SIFSIX-1-Cu displays moderately high C2H2/C2H4 separation selectivities
(7.1 to 10.6), which are greater than those of
FeMOF-74 (2.1) (18) and NOTT-300 (2.2 to 2.5)
(14) (Fig. 2A). For C2H2/C2H4 (1/99) mixtures,
SIFSIX-1-Cu exhibits greater selectivities (10.6)
than SIFSIX-3-Zn (8.8) and SIFSIX-3-Ni (5.0)
(Fig. 2B). Its C2H2 uptake from the 1/99 mixture

is the second highest of the compounds investigated (Fig. 2D). Given that SIFSIX-1-Cu has a
relatively large surface area, it should also be
very efficient for C2H2/C2H4 (1/99) separation.
SIFSIX-1-Cu exhibits higher C2H2 uptakes than
benchmark MOFs, including FeMOF-74, NOTT300, and UTSA-100a. Its high C2H2/C2H4 selectivities (Fig. 2A) make SIFSIX-1-Cu most suitable
for C2H2/C2H4 (50/50) separation (Fig. 2C and
fig. S20) (24). Some further comparisons of how
these MOFs perform with respect to C2H2/C2H4
separations are shown in table S1 (24).
Transient breakthrough simulations (27) were
conducted to demonstrate the C2H2/C2H4 separation performances of the SIFSIX materials in
column adsorption processes. Two C2H2/C2H4 mixtures (1/99 and 50/50) were used as feeds to mimic
the industrial process conditions. Clean separations
were realized with all five SIFSIX MOFs; C2H4 first
eluted through the bed to yield a polymer-grade gas,
then C2H2 broke through from the bed at a certain
time tbreak (Fig. 3A and figs. S21 and S22) (24). The
dimensionless tbreak values for SIFSIX-1-Cu (1/99 and
50/50 mixtures) and SIFSIX-2-Cu-i (1/99 mixture)
exceed those of the other SIFSIX materials and
MOFs that we studied (Fig. 3, A and B). The amount
of C2H2 captured from the 1/99 mixture in SIFSIX1-Cu and SIFSIX-2-Cu-i was as high as 265.3 and
780.0 mmol/liter, respectively, which compares
favorably with state-of-art adsorbents such as UTSA100a (135.5 mmol/liter), FeMOF-74 (100.7 mmol/liter),
and NOTT-300 (68.3 mmol/liter). SIFSIX-2-Cu-i
efficiently removes trace C2H2 from C2H4 gas
(1/99 mixture), whereas SIFSIX-1-Cu demonstrates excellent C2H2 capacity with an uptake

of 5533 mmol/liter from the 50/50 mixture,

~37% greater than that of FeMOF-74 (Fig. 3F).
Through experimental breakthrough studies,
we further examined these materials in actual
adsorption processes for both 1/99 and 50/50 mixtures. Highly efficient separations for C2H2/C2H4
mixtures were realized (Fig. 3, D and E). For the
capture of C2H2 from the 1/99 mixture, the concentration of C2H2 in the gas exiting the adsorber
for up to 140 min was below 2 ppm, and the purity of C2H4 was >99.998% (fig. S23) (24). The
hierarchy of breakthrough times for the 1/99 mixture was, from longest to shortest, SIFSIX-2-Cu-i,
SIFSIX-1-Cu, then SIFSIX-3-Zn; for the 50/50 mixture, it was SIFSIX-1-Cu, SIFSIX-3-Zn, then SIFSIX2-Cu-i (Fig. 3, D and E). These experiments are
consistent with simulated breakthrough results.
Although the uptake of C2H2 by SIFSIX-3-Zn at low
pressure for the 1/99 mixture is higher than that by
SIFSIX-1-Cu (Fig. 1B), SIFSIX-1-Cu exhibits a longer
breakthrough time for C2H2, presumably because of
its higher selectivity (Fig. 2B). The amounts of
C2H2 captured by SIFSIX-1-Cu, SIFSIX-2-Cu-i, and
SIFSIX-3-Zn from the 1/99 mixture (0.38, 0.73, and
0.08 mmol/g, respectively) and from the 50/50
mixture (6.37, 2.88, and 1.52 mmol/g, respectively)
during the breakthrough process are in excellent
agreement with the simulated results, except in
the case of SIFSIX-3-Zn (tables S12 and S13) (24).
In the production of high-purity C2H4, the feed
gases for the C2H2 removal unit are contaminated
with trace levels of CO2 (<50 ppm), H2O (<5 ppm),
and O2 (<5 ppm). We conducted breakthrough experiments for the 1/99 mixture with SIFSIX-2-Cu-i,
the best-performing material for capturing trace
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




1. S. Kitagawa, Angew. Chem. Int. Ed. 54, 1068610687

2. R. Vaidhyanathan et al., Science 330, 650653 (2010).
3. N. T. T. Nguyen et al., Angew. Chem. Int. Ed. 53, 1064510648
4. N. L. Rosi et al., Science 300, 11271129 (2003).
5. Y. Peng et al., J. Am. Chem. Soc. 135, 1188711894 (2013).
6. J. A. Mason et al., Nature 527, 357361 (2015).
7. S. Ma et al., J. Am. Chem. Soc. 130, 10121016 (2008).
8. H. Furukawa, K. E. Cordova, M. OKeeffe, O. M. Yaghi, Science
341, 1230444 (2013).
9. S. J. Datta et al., Science 350, 302306 (2015).
10. Z. J. Zhang, Z.-Z. Yao, S. Xiang, B. Chen, Energy Environ. Sci. 7,
28682899 (2014).
11. Q. Lin, T. Wu, S. T. Zheng, X. Bu, P. Feng, J. Am. Chem. Soc.
134, 784787 (2012).
12. J.-R. Li, R. J. Kuppler, H.-C. Zhou, Chem. Soc. Rev. 38,
14771504 (2009).
13. R. Matsuda et al., Nature 436, 238241 (2005).
14. S. Yang et al., Nat. Chem. 7, 121129 (2014).
15. H. Wu, Q. Gong, D. H. Olson, J. Li, Chem. Rev. 112, 836868
16. J.-P. Zhang, X.-M. Chen, J. Am. Chem. Soc. 131, 55165521
17. T. L. Hu et al., Nat. Commun. 6, 73287335 (2015).
18. E. D. Bloch et al., Science 335, 16061610 (2012).
19. M. C. Das et al., J. Am. Chem. Soc. 134, 87038710 (2012).
20. P. Nugent et al., Nature 495, 8084 (2013).
21. S. Noro et al., J. Am. Chem. Soc. 124, 25682583 (2002).
22. O. Shekhah et al., Nat. Commun. 5, 42284234 (2014).
23. A. Kumar et al., Angew. Chem. Int. Ed. 54, 1437214377
24. Materials and methods are available as supplementary
materials on Science Online.
25. F. Studt et al., Science 320, 13201322 (2008).
26. K. S. Walton, D. S. Sholl, AIChE J. 61, 27572762 (2015).
27. R. Krishna, RSC Adv. 5, 5226952295 (2015).

Zhejiang Provincial Natural Science Foundation of China

(grant LR13B060001), Ten Thousand Talent Program of China
(to H.X.), the Welch Foundation (grant AX-1730), King Abdullah
Science and Technology University Office of Competitive
Research Funds (grant URF/1/1672-01-01), and the Science
Foundation Ireland (award 13/RP/B2549 to M.Z.). We thank
T. L. Hu, Y. F. Zhao, W. D. Shan, and M. D. Jiang for their help and
arrangement of the breakthrough experiments; A. Kumar for
help with sample characterization; and Z. G. Zhang and B. G. Su for
discussions of the experiments. Metrical data for the solid-state
structures of SIFSIX-2-Cu-iC2D2 and SIFSIX-1-CuC2D2 are
available free of charge from the Cambridge Crystallographic Data
Centre under reference numbers CCDC 1471795 and 1471796.
Materials and Methods
Figs. S1 to S32
Tables S1 to S15
References (2834)
12 January 2016; accepted 5 May 2016
Published online 19 May 2016

Copper-catalyzed asymmetric
addition of olefin-derived
nucleophiles to ketones
Yang Yang,1 Ian B. Perry,1 Gang Lu,2 Peng Liu,2* Stephen L. Buchwald1*
Enantioenriched alcohols found in an array of bioactive natural products and
pharmaceutical agents are often synthesized by asymmetric nucleophilic addition to
carbonyls. However, this approach generally shows limited functional-group compatibility,
requiring the use of preformed organometallic reagents in conjunction with a
stoichiometric or substoichiometric amount of chiral controller to deliver optically
active alcohols. Herein we report a copper-catalyzed strategy for the stereoselective
nucleophilic addition of propargylic and other alkyl groups to ketones, using easily
accessible (poly)unsaturated hydrocarbons as latent carbanion equivalents. Our method
features the catalytic generation of highly enantioenriched organocopper intermediates
and their subsequent diastereoselective addition to ketones, allowing for the effective
construction of highly substituted stereochemical dyads with excellent stereocontrol.
Moreover, this process is general, scalable, and occurs at ambient temperature.

tereochemically complex alcohols in optically pure form are commonly encountered

structural elements in a diverse range of
pharmaceutical drugs and biologically active
natural products (Fig. 1A). Consequently,
general methods that allow for the stereoselective
assembly of highly substituted alcohols have long
been sought (1). The discovery of Grignard reagents
and their subsequent addition to ketones and
aldehydes have been widely considered as milestones
in synthetic chemistry, giving rise to a general
synthesis of alcohols from preformed organomagnesium reagents and broadly available carbonyl compounds (2). Since then, extensive efforts
have been devoted to the development of asymmetric variants of nucleophilic addition reactions
to carbonyls, using preformed organometallic reagents (1, 37). These synthetic endeavors have
proven to be exceptionally fruitful, culminating
in a variety of protocols for enantioselective
additions to carbonyls, using either a chiral auxiliary-


This work is supported by the National Natural Science

Foundation of China (grants 21222601, 21436010, and 21476192),

*Corresponding author. Email: (P.L.); (S.L.B.)

8 JULY 2016 VOL 353 ISSUE 6295



Department of Chemistry, Massachusetts Institute of

Technology (MIT), Cambridge, MA 02139, USA. 2Department
of Chemistry, University of Pittsburgh, Pittsburgh, PA 15260,


The authors and their affiliated institutions have filed a patent

application related to the results presented here.

modified organometallic reagent or a substoichiometric amount of chiral controller to achieve

excellent levels of stereocontrol. Compared with
aldehydes, however, the asymmetric nucleophilic
addition to ketones has been studied to a lesser
extent (57).
Although numerous advances have been made
in this area, considerable hurdles have impeded
the further adaptation of these methods by the
synthetic community. The requirement to prepare and use a stoichiometric quantity of an
organometallic reagent complicates most of the
existing methods. In addition, the highly nucleophilic and basic nature of organometallic reagents has posed substantial limitations with
respect to the functional-group compatibility of
these processes. As a consequence, these methods
are typically not amenable to the transformation
of late-stage intermediates and other highly functionalized molecules. Furthermore, an additional
synthetic operation is required to prepare these
organometallic reagents from organic halide or
unsaturated hydrocarbon precursors, imposing
further constraints on the types of nucleophiles
suitable for carbonyl addition.
In this context, a catalytic method for stereoselective additions to carbonyls, using easily accessible olefins as latent carbanion equivalents in SCIENCE

Downloaded from on July 8, 2016

amounts of C2H2. These experiments indicate that

the presence of CO2 has only a slight (1000 ppm
CO2) or no (10 ppm CO2) effect on the separation
of C2H2 from C2H4 (fig. S24) (24). Moisture (6 to
1340 ppm) and oxygen (2200 ppm) do not affect
the C2H2 capture ability of SIFSIX-2-Cu-i (figs. S25
and S26) (24). The breakthrough performances of
SIFSIX-2-Cu-i and SIFSIX-3-Zn for the 1/99 mixture did not decline during 16 and 3 cycles, respectively (figs. S28 and S29) (24), and the SIFSIX
materials retained their stability after breakthrough experiments (figs. S1 and S30) (24).
The SIFSIX materials that we studied exhibit
excellent C2H2 storage performance. The volumetric uptake of C2H2 by SIFSIX-1-Cu at 298 K and
1.0 bar is the highest among these SIFSIX materials
(0.191 g/cm3; table S14) (24). The C2H2 storage densities in the pores of SIFSIX-1-Cu, SIFSIX-2-Cu-i,
and SIFSIX-3-Zn at 298 K are 0.388, 0.403, and
0.499 g/cm3, respectively.
The basic principles outlined here are likely to
be applicable to other gas mixtures. Primary binding
sites will be necessary for recognition of specific gas
molecules, whereas suitable pore sizes and spacing
will be needed to enforce synergistic binding to
multiple sites in order to form the so-called gas
clusters through intermolecular guest-guest interactions. This work not only reveals a path
forward for industrial C2H2/C2H4 separations,
but also facilitates a design or crystal engineering
approach to the development of porous materials for other gas separations.


remains challenging with these processes, in part

because of the increased steric hindrance and the
attenuated electrophilicity of ketones relative
to aldehydes. In a mechanistically distinct approach, Montgomery, Jamison, and colleagues
have devised novel Ni-based methods for the enantioselective reductive coupling of alkynes and
aldehydes (1114). Additionally, Hoveyda and
co-workers have developed exceptional coppercatalyzed borylative processes for the addition
of olefin-derived nucleophiles bearing a boryl moiety to carbonyls. This chemistry has been highly

successful in the enantioselective preparation

of secondary alcohols from aldehydes (15, 16),
although ketones represent more challenging
substrates for this enantioselective process (16).
Alkenes and ketones constitute two of the most
widely available building blocks for organic synthesis. Nevertheless, a broadly applicable and
highly enantioselective catalyst system for the
coupling of these orthogonal feedstocks remains
We have recently become interested in the
development of copper(I) hydride (CuH) catalysis

Downloaded from on July 8, 2016

lieu of preformed organometallic reagents, would

circumvent these issues and greatly expand the
scope of stereochemically well-defined alcohols
accessible from carbonyl addition processes. In
elegant work using Ru-, Rh-, and Ir-based noble
metal catalysts, Krische and co-workers have
pioneered the development of a range of synthetically versatile stereoselective coupling reactions
of olefins and carbonyls under hydrogenation
and transfer hydrogenation conditions (810).
To date, however, the synthesis of sterically demanding tertiary alcohols from simple ketones

Fig. 1. (A to D) Asymmetric addition of olefin-derived nucleophiles to ketones, using a copper-hydride catalyst. M, metal; L, ligand; R1 to R4, generic
substituent; tBu, tert-butyl; Ph, phenyl; RT, room temperature; krel, relative rate constant.

8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 2. Substrate scope of the copper-catalyzed diastereo- and enantioselective addition of enyne-derived nucleophiles to ketones. Reaction
conditions: Cu(OAc)2 (0.5 mol %), (S,S)-Ph-BPE (6 mol %), ketone [0.5 mmol,
1.0 equivalents (equiv)], enyne (0.75 mmol, 1.5 equiv), t-BuOH (0.5 mmol,
1 equiv), (MeO)2MeSiH (2.5 mmol, 5 equiv), RT, 12 hours. Yields refer to isolated
yields of the major diastereomer (>10:1 dr). Yields in parentheses refer to the


8 JULY 2016 VOL 353 ISSUE 6295

combined yields of two diastereomers determined by 1H nuclear magnetic

resonance (NMR) spectroscopy calibrated using an internal standard. The dr
values were determined by 1H NMR spectroscopic analysis of the crude reaction
mixture, and the ee values were determined by chiral high-performance liquid
chromatography analysis. TBS, tert-butyldimethylsilyl; Ts, tosyl; Et, ethyl. Blue
shading highlights the functional group present in the starting material. SCIENCE

Fig. 3. Application and further extension of the copper-catalyzed asymmetric addition of olefinderived nucleophiles. (A) a: Pd(OH)2/C, H2 (balloon pressure), MeOH, RT, 12 hours. b: 5 mol % IPrCuCl,
6 mol % LiOtBu, (MeO)2MeSiH, t-BuOH, toluene, RT, 48 hours. (B) c: 5 mol % Cu(OAc)2, 6 mol % (S,S)-PhBPE, (MeO)2MeSiH, t-BuOH, cyclohexane, RT, 12 hours. (C) d: 5 mol % Cu(OAc)2, 6 mol % (S,S)-Ph-BPE,
17 (2.0 equiv), 1 (1.0 equiv), (MeO)2MeSiH, tetrahydrofuran, RT, 12 hours. e: 5 mol % Cu(OAc)2, 6 mol % (S,S)Ph-BPE, 19 (2.0 equiv), 20 (1.0 equiv), (MeO)2MeSiH, methyl tert-butyl ether, RT, 12 hours. f: 5 mol %
Cu(OAc)2, 6 mol % (S,S)-Ph-BPE, 3 (1.5 equiv), 22 (1.0 equiv), (MeO)2MeSiH, t-BuOH, cyclohexane,
RT, 12 hours.


as a general means of accessing enantioenriched

alkylcopper intermediates from readily available
olefin precursors. By employing a class of electrophilic aminating reagents established by Berman
and Johnson (17) to intercept these organometallic intermediates, our laboratory (18) and
Miura, Hirano, and colleagues (1921) have independently developed a CuH-catalyzed approach
for the enantioselective hydroamination of olefins
(22). On the basis of these studies and notable
contributions from other groups in the area of
copper-catalyzed enantioselective reductive aldol
reactions (6, 7, 2326), we envisioned the development of CuH chemistry as a general platform
for the use of simple olefins as latent carbon nucleophiles in carbonyl addition (Fig. 1B). In particular, we were interested in investigating the
coupling of (poly)unsaturated hydrocarbons and
ketones via the diastereoselective addition of
catalytically generated chiral organocopper intermediates to the carbonyl group. If successful, our
proposed strategy would enable the use of an
earth-abundant first-row transition metal catalyst
for the diastereo- and enantioselective synthesis
of alcohols, including those possessing vicinal triand tetrasubstituted stereogenic centers from
olefins and carbonyls.
Our proposed catalytic cycle is illustrated in
Fig. 1C. Enantioselective addition of a phosphinebound copper hydride catalyst (I) across the C=C
double bond of an olefin (II) would lead to the
formation of an alkylcopper intermediate (III).
Interception of this nascent alkylcopper intermediate with a ketone (IV) would provide the
alkoxycopper species V. Ligand exchange of V
with tert-butanol (VI) would furnish the asymmetric addition product (VII) and release copper
tert-butoxide (VIII). Upon undergoing a s-bond
metathesis with hydrosilane IX, VIII would
then regenerate the copper hydride catalyst (I)
and complete the catalytic cycle. The mechanistic
proposal outlined herein does not require the use
of any exogenous acidic or basic additives for
substrate or catalyst activation, thereby potentially allowing for a range of sensitive functional
groups to be tolerated. Mechanistically, in contrast
to the majority of previously developed transition
metalcatalyzed reductive couplings (814), the
strategy we envisioned does not involve redox
processes at the copper(I) center while allowing for the polar and nonpolar building blocks
to couple in a reductive fashion. Additionally,
inexpensive and mild silane reagents are used
as the reductant for the catalytic generation of
organocopper intermediates.
At the outset, we recognized that the successful
implementation of our proposed strategy would
depend on unprecedented chemo- and stereoselectivity with respect to the copper catalyst.
The competitive hydrocupration of the carbonyl
coupling partner is known to be a facile process
and would lead to its unproductive reduction.
Moreover, important work by Lipshutz and coworkers has demonstrated that the copper hydride species based on DTBM-SEGPHOS [DTBM,
3,5-di-tert-butyl-4-methoxyphenyl; SEGPHOS, 4,4bi-1,3-benzodioxole-5,5-diylbis(diphenylphosphane)]
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



We focused our initial investigation on the

transformation of enynes, a subset of olefins that
could be conveniently prepared from terminal
alkynes in a single operation. Previous work on
the use of enynes for transition metalcatalyzed
reductive and borylative transformations (15, 2830)
led us to postulate that these enyne substrates
could serve as a suitable starting point for our
investigations, because hydrocupration of these
substrates might be facilitated by the formation
of an alkylcopper species stabilized by a neighboring alkynyl group. From the standpoint of
utility, the enantioselective coupling of enynes
to ketones would furnish synthetically versatile
homopropargyl alcohol products (3133). Prior
to our experimental studies, we carried out density functional theory (DFT) calculations to gain
some insight into the ligand effects on the ketone
reduction and olefin hydrocupration processes

(Fig. 1D). Our computational studies revealed

that, although the hydrocupration of ketones is
a kinetically facile and highly exergonic process,
the choice of the ancillary phosphine ligand could
dramatically influence the reactivity of the L*CuH
species. In particular, although the reduction of
ketone 1 with a catalyst generated from DTBMSEGPHOS requires a low activation energy (DG)
of 9.9 kcal/mol, the same process is less facile
when other commercially available ligands are
employed. In particular, the use of a catalyst
derived from Ph-BPE {(S,S)-Ph-BPE, ()-1,2bis[(2S,5S)-2,5-diphenylphospholano]ethane} (L2),
another ligand previously employed in CuH chemistry (20, 34), results in a higher activation energy
(DG = 13.6 kcal/mol). Furthermore, our calculations suggested that the activation barrier for the
hydrocupration of enyne 3 is 1.3 kcal/mol lower
than that of ketone 1 with this Ph-BPEbased

Downloaded from on July 8, 2016

(L1), our optimal catalyst for olefin hydroamination (18), is highly robust and efficient for the
asymmetric reduction of a variety of carbonyls,
including ketones (27). On the other hand, the
alkylcopper intermediate generated from olefin
hydrometalation would need to be nucleophilic
enough to facilitate the alkylation of a range of
unactivated ketones, a process that has proven
challenging with other transition metal catalysts.
Nevertheless, we reasoned that judicious choice
of the copper catalyst could suppress the undesired ketone reduction while still allowing for
the effective olefin hydrocupration, as well as
ketone alkylation. Furthermore, we postulated
that it would be feasible to access a catalyst capable of imposing effective stereochemical facial
discrimination on the prochiral olefin and the
ketone to achieve high levels of enantio- and

Fig. 4. DFT-calculated enantio- and diastereochemicaldetermining transition states for the copper-catalyzed addition of an enyne (12)derived
nucleophile to ketone 1. (A to C) Energies were calculated at the M06/SDD6-311+G(d,p)/SMD(cyclohexane) level of theory with geometries optimized at the
B3LYP/SDD6-31G(d) level. DGsol, free energy of activation; DHsol, enthalpy of activation. In (B), roman numerals denote the four quadrants of the coordination space;
green indicates the enyne substrate; yellow indicates the hydride of LCuH. In (C), green indicates the chiral allenyl moiety; yellow indicates a six-membered ring.


8 JULY 2016 VOL 353 ISSUE 6295 SCIENCE

catalyst, indicating the feasibility of chemoselective enyne hydrocupration in the presence of

ketones. These computational trends were corroborated by our experimental efforts to identify suitable conditions for the enantioselective
addition of enyne-derived nucleophiles to ketones.
The use of a catalyst based on DTBM-SEGPHOS
produced the coupled product 4a in only 48%
yield with moderate stereoselectivity [2.3:1 diastereomeric ratio (dr), 85% enantiomeric excess
(ee)], along with a substantial amount of reduction product 4b (52% yield). In contrast, under
the optimized reaction conditions employing
Ph-BPE as the ligand, the coupled product bearing
adjacent tertiary and quaternary stereocenters
(4a) formed in 97% yield with excellent diastereoand enantiocontrol (>20:1 dr, 99.5% ee).
We next set out to examine the substrate scope
of this enyne-ketone coupling process (Fig. 2).
Electron-rich and electron-deficient acetophenones were successfully transformed into the
corresponding tertiary alcohols (5b and 5c) in
good yield with excellent diastereo- and enantioselectivity. Ketones bearing aryl bromide (5d)
and chloride (5e) functional groups were also
compatible with our protocol, permitting further
transformations of the alcohol product using crosscoupling technologies. Additionally, a variety of
nitrogen heterocycles that are frequently found
in medicinally active agents, including a pyrrole
(5h), a thiazole (5i), a pyrazole (5j), a pyridine
(5k), a thiophene (5x), and a carbazole (6a), could
be applied with this stereoselective coupling. Appreciably less reactive aliphatic ketones could also
be effectively converted into tertiary aliphatic alcohols with excellent enantioselectivity (5l to 5o),
further highlighting the excellent reactivity exhibited by this Ph-BPEbased alkylcopper intermediate. An a-ketoester (5q) and an enone (5r)
also underwent the asymmetric carbonyl addition
with outstanding enantiocontrol. Moreover, the
current process displayed a broad scope with respect to the enyne fragment. In addition to enynes
bearing aromatic substituents, aliphatic enynes
were also readily accommodated under these reaction conditions (5h, 5i, 5r to 5v, and 6a to 6e).
Finally, substitution at the olefinic terminus of
the enyne component (5y) was found to be compatible with our protocol.
The mildness of the reaction conditions required for the current asymmetric nucleophilic
addition allowed for the use of substrates possessing sensitive functional groups that are incompatible with preformed organometallic reagents.
A variety of sensitive functional groups, such as a
thioether (5f), a sulfonamide (5t), a methyl ester
(5q), an alkyl tosylate (5u), an alkyl chloride
(5v), and a tertiary amine (5w), were tolerated
under these reaction conditions. Most usefully,
various basic and acidic functional groups, including a secondary amine (6a), a secondary
amide (6b), an alcohol (6c), a phenol (6d), and
a carboxylic acid (6e), were compatible with our
protocol, thereby eliminating the need for inefficient protection and deprotection sequences often
required by traditional carbonyl alkylation methodologies that use stoichiometric organometallic

reagents. The relative and absolute stereochemistry of the addition product were determined
by x-ray crystallographic analysis of 5f.
In addition, this copper-catalyzed enantioselective coupling process proved to be scalable
and required a very low catalyst loading (Fig. 3A).
We were able to perform a 50-mmol-scale (13-gscale) reaction to prepare 5c, using only 0.2 mole %
(mol %) Cu(OAc)2 (OAc, acetate) and 0.4 mol %
(S,S)-Ph-BPE without lowering the yield or the
stereoselectivity of the transformation (82% yield,
10:1 dr, 98% ee). The scalability and robustness of
this copper-catalyzed asymmetric transformation
illustrate its potential application in the manufacture of highly enantioenriched fine chemicals on
an industrial scale. Furthermore, these synthetically versatile homopropargyl alcohols prepared
by our method could easily be transformed into a
variety of useful molecular architectures. For
example, hydrogenation and CuH-catalyzed cisselective semireduction of alkyne 5c (35) furnished saturated alcohol 9a and homoallylic
alcohol 9b as a single stereoisomer, respectively.
These reactions allowed for the synthetic equivalent of asymmetric addition to ketones with
nucleophiles derived from terminal aliphatic
olefins and 1,3-dienes in a highly regio-, diastereo-,
and enantioselective fashion.
The late-stage modification of medicinally
relevant molecules allows for the rapid generation
of analogous bioactive compounds and facilitates
structure-activity relationship studies. To further
demonstrate the synthetic utility of our method
in this context, we selected two widely prescribed
pharmaceutical agents, terbinafine (Lamisil, 10a)
and fenofibrate (Tricor, 10b) and subjected them
to our coupling protocol (Fig. 3B). Both substrates
were effectively converted into the tertiary alcohol
adduct with excellent levels of enantioselectivity
(99% ee for each), further highlighting the utility
of our method in a more complex setting.
We also performed DFT calculations to investigate the mechanism and the origin of stereoselectivity of this copper-catalyzed enyne-ketone
coupling. The computed free energy profile of
the overall catalytic cycle is shown in Fig. 4A (see
supplementary materials for a detailed discussion). Our computational studies suggest that
the initial enyne hydrocupration is irreversible
and thus constitutes the enantio-determining
step for this reaction. The hydrocupration involves a four-membered transition state in which
the delivery of the hydride and the formation of
the CuC bond take place simultaneously. To
better understand the enantioinduction that occurs during the hydrocupration event, we divided
the coordination space of this C2 symmetric copper catalyst into four quadrants (Fig. 4B). The
phenyl substituents on the ligands phospholane
moieties in quadrants II and IV point out of the
plane of the paper, rendering these two quadrants sterically inaccessible for a substrate coming
toward this plane. In contrast, quadrants I and
III are relatively unencumbered, with the two
phenyl groups pointing away from the plane of
the paper. In TS2a, the unfavorable repulsion
between the olefins alkynyl substituent and the

catalysts sterically crowded quadrant IV is avoided,

thereby leading to the preferential formation of the
(S)-propargylcopper intermediate 13a, with an
activation free energy that is 5.5 kcal/mol lower
than the disfavored transition state TS2b.
Subsequent 1,3-isomerization of the highly
enantioenriched propargylcopper 13a proceeds
in a stereospecific fashion, affording an axially
chiral allenylcopper intermediate 14 as a single
diastereomer. Furthermore, this 1,3 shift is found
to be facile and exergonic (DG = 2.6 kcal/mol),
thus setting the stage for the ketone addition
involving a six-membered cyclic transition state
(Fig. 4C). In the lowest-energy transition states of
this diastereoselectivity-determining step (TS4a
and TS4b), the phenyl group of the ketone is
oriented anti to the allenylcopper [i.e., gauche
to the methyl (Me) and H substituents on the
distal terminus of the allenyl moiety] to avoid
close contact with the Ph-PBE ligand (see supplementary materials for details). In the preferred
diastereomeric transition state TS4a, the chiral
allenyl moiety (highlighted in green) attacks the
(Re)-face of the ketone, placing the hydrogen
atom of the allene gauche to the methyl and the
phenyl groups flanking the carbonyl group. In
the disfavored transition state TS4b [(Si)-face attack], the bulkier Me group of the allene is
oriented gauche to the two carbonyl substituents.
In addition, TS4b is also destabilized by the
proximity of the allenyl moiety with the Ph-BPE
ligand, as evidenced by a relatively short HH
distance of 2.32 . Taken together, the substratesubstrate and ligand-substrate interactions combine to confer high levels of diastereocontrol in
the CC bond forming event leading to the
homopropargyl copper alkoxide (15a) and, eventually, the homopropargyl alcohol product (16).
In preliminary studies, we have demonstrated
that a range of other structurally diversified olefins
could also be employed in this mode of catalysis
in a highly enantioselective fashion (Fig. 3C). By
applying slightly modified reaction conditions, a
related process with a cyclic diene (17) and acetophenone (1) afforded tertiary alcohols possessing
contiguous tertiary and quaternary stereocenters
in >10:1 dr and 94% ee. Styrenes (19), a class of
abundant feedstock chemicals possessing moderate levels of reactivity toward hydrometalation,
were also effectively converted into the tertiary
alcohol adduct with excellent enantiocontrol
(95% ee). Furthermore, thione 22, which could
easily be synthesized from the parent ketone, was
also suitable for this coupling process. The highly
enantioselective preparation of tertiary thiol 23
(>10:1 dr, 96% ee) using our strategy represents a
synthetically valuable advance, given that enantioenriched thiols are prevalent structural motifs
in a variety of small-molecule therapeutics, and
only a handful of enantioselective methods are
available for their preparation (36).
Our strategy represents a departure from the
use of preformed organometallic reagents in
classic carbonyl addition chemistry. Furthermore,
this CC bond forming process readily accommodates a wide range of olefin and ketone coupling
partners, allowing for the highly stereoselective
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



preparation of alcohols bearing adjacent tri- and

tetrasubstituted stereogenic centers. We anticipate that the design of optimal ligand scaffolds
will enable the transformation of a broader range
of unactivated olefins, ultimately delivering a
powerful tool for the asymmetric addition of
olefin-derived nucleophiles to carbonyls that will
be of broad synthetic utility.

crystallographic data for 5f. Calculations were performed at the

Center for Simulation and Modeling at the University of Pittsburgh
and the Extreme Science and Engineering Discovery Environment
(XSEDE) supported by the NSF.

Table S1
References (3750)
NMR Spectra
HPLC Traces

Materials and Methods
Figs. S1 to S12

28 March 2016; accepted 20 May 2016

Published online 9 June 2016



We are grateful to the NIH (grant GM 46059 to S.L.B.), the

Undergraduate Research Opportunities Program at MIT (I.B.P.),
and the University of Pittsburgh (P.L.) for financial support.
The content in this paper is solely the responsibility of the authors
and does not necessarily represent the official views of the NIH.
We thank Y.-M. Wang and M. T. Pirnot (MIT) for advice on the
preparation of this manuscript and P. Mller (MIT) for x-ray
crystallographic analysis of 5f. Cambridge Crystallographic Data
Centre accession number 1470340 contains the supplementary


8 JULY 2016 VOL 353 ISSUE 6295


Thermally stable single-atom

platinum-on-ceria catalysts
via atom trapping
John Jones,1* Haifeng Xiong,1* Andrew T. DeLaRiva,1 Eric J. Peterson,1 Hien Pham,1
Sivakumar R. Challa,1 Gongshin Qi,2 Se Oh,2 Michelle H. Wiebenga,2
Xavier Isidro Pereira Hernndez,3 Yong Wang,3,4 Abhaya K. Datye1
Catalysts based on single atoms of scarce precious metals can lead to more efficient use
through enhanced reactivity and selectivity. However, single atoms on catalyst supports
can be mobile and aggregate into nanoparticles when heated at elevated temperatures.
High temperatures are detrimental to catalyst performance unless these mobile atoms can
be trapped. We used ceria powders having similar surface areas but different exposed
surface facets. When mixed with a platinum/aluminum oxide catalyst and aged in air at
800C, the platinum transferred to the ceria and was trapped. Polyhedral ceria and
nanorods were more effective than ceria cubes at anchoring the platinum. Performing
synthesis at high temperatures ensures that only the most stable binding sites are
occupied, yielding a sinter-resistant, atomically dispersed catalyst.

iesel oxidation catalysts (DOCs) are used

in the automotive industry to oxidize harmful emissions that contain CO, NO, and
hydrocarbons. Pt is very active for such
oxidation reactions (1, 2). Under oxidizing
conditions at high temperatures, the Pt nanoparticles on the oxide support sinter to form
large particles, which leads to a loss of surface
area and hence decreased catalytic activity, so
there is commercial interest in developing sinterresistant catalysts that can maintain activity over
long-term operation. We previously found that
the mechanism of activity loss involves the emission of mobile species from small particles and
their capture by large particlesa process called
Ostwald ripening (3, 4). Here, we show that the
conditions under which nanoparticles emit mobile species are also ideally suited to generating
atomically dispersed catalysts if the mobile species can be effectively trapped. Platinum group
metals can be trapped in ionic form in complex

Department of Chemical and Biological Engineering and

Center for Micro-Engineered Materials, University of New
Mexico, Albuquerque, NM 87131, USA. 2General Motors
Global R&D, 30500 Mound Road, Warren, MI 48090, USA.
Voiland School of Chemical Engineering and Bioengineering,
Washington State University, Pullman, WA 99164, USA.
Institute for Integrated Catalysis, Pacific Northwest National
Laboratory, Richland, WA 99354, USA.
*These authors contributed equally to this work. Corresponding
author. Email:

oxides (5), but they can be lost inside the bulk

oxide, where they are no longer effective as a
catalyst. An ideal support for trapping metal species is one in which the mobile species stay at
the surface so that they retain catalytic activity.
Our studies were performed at 800C in
oxidizing ambient conditions typically used for
accelerated testing of DOCs. Under these conditions, Pt can be emitted as volatile PtO2 (6)
which can lead to transport of Pt from the DOC
catalyst to the downstream selective catalytic
reduction (SCR) catalyst (7). A vapor-phase approach for the synthesis of catalystsspecifically,
transferring Pt from a metal foil to aluminawas
reported previously (8), but the ability of supports such as ceria to trap atomically dispersed
Pt was not recognized. In industry, the vaporphase transport of metal oxides is regarded as a
leading cause of catalyst degradation (9). Thus,
we set out to find materials that could effectively trap the mobile species. In previous work,
we found that PdO was able to trap mobile PtO2,
forming metallic Pt-Pd particles (10). The remarkable efficiency of PdO for trapping Pt suggests that other oxides could play a similar role.
McVicker et al. (11) had implied that CaO, SrO, and
BaO may slow the rates of Ir sintering through
the presence of acid sites on the support.
We worked with ceria as a support because
Ce-Zr-Y was shown by Nagai et al. (12) to be
effective for inhibiting the sintering of Pt by SCIENCE

Downloaded from on July 8, 2016

1. E. J. Corey, L. Krti, Enantioselective Chemical Synthesis:

Methods, Logic and Practice (Direct Book Publishing, 2010).
2. P. Knochel et al., Angew. Chem. Int. Ed. 42, 43024320 (2003).
3. R. Noyori, M. Kitamura, Angew. Chem. Int. Ed. Engl. 30, 4969
4. L. Pu, H.-B. Yu, Chem. Rev. 101, 757824 (2001).
5. J. L. Stymiest, V. Bagutski, R. M. French, V. K. Aggarwal,
Nature 456, 778782 (2008).
6. O. Riant, J. Hannedouche, Org. Biomol. Chem. 5, 873888 (2007).
7. M. Shibasaki, M. Kanai, Chem. Rev. 108, 28532873 (2008).
8. H.-Y. Jang, M. J. Krische, Acc. Chem. Res. 37, 653661 (2004).
9. E. Skucas, M.-Y. Ngai, V. Komanduri, M. J. Krische, Acc. Chem.
Res. 40, 13941401 (2007).
10. J. F. Bower, I. S. Kim, R. L. Patman, M. J. Krische, Angew.
Chem. Int. Ed. 48, 3446 (2009).
11. M. R. Chaulagain, G. J. Sormunen, J. Montgomery, J. Am.
Chem. Soc. 129, 95689569 (2007).
12. E. P. Jackson et al., Acc. Chem. Res. 48, 17361745 (2015).
13. K. M. Miller, W.-S. Huang, T. F. Jamison, J. Am. Chem. Soc. 125,
34423443 (2004).
14. R. M. Moslin, K. Miller-Moslin, T. F. Jamison, Chem. Commun.
2007, 44414449 (2007).
15. F. Meng, F. Haeffner, A. H. Hoveyda, J. Am. Chem. Soc. 136,
1130411307 (2014).
16. F. Meng, H. Jang, B. Jung, A. H. Hoveyda, Angew. Chem. Int. Ed.
52, 50465051 (2013).
17. A. M. Berman, J. S. Johnson, J. Am. Chem. Soc. 126, 56805681 (2004).
18. M. T. Pirnot, Y.-M. Wang, S. L. Buchwald, Angew. Chem. Int. Ed.
55, 4857 (2016).
19. Y. Mikki, K. Hirano, T. Satoh, M. Miura, Angew. Chem. Int. Ed.
52, 1083010834 (2013).
20. Y. Mikki, K. Hirano, T. Satoh, M. Miura, Org. Lett. 16, 14981501
21. D. Nishikawa, K. Hirano, M. Miura, J. Am. Chem. Soc. 137,
1562015623 (2015).
22. Y. Xi, T. W. Butcher, J. Zhang, J. F. Hartwig, Angew. Chem. Int.
Ed. 55, 776780 (2016).
23. J. Deschamp, O. Chuzel, J. Hannedouche, O. Riant, Angew.
Chem. Int. Ed. 45, 12921297 (2006).
24. D. Zhao, K. Oisaki, M. Kanai, M. Shibasaki, J. Am. Chem. Soc.
128, 1444014441 (2006).
25. B. H. Lipshutz, B. Amorelli, J. B. Unger, J. Am. Chem. Soc. 130,
1437814379 (2008).
26. A. Saxena, B. Choi, H. W. Lam, J. Am. Chem. Soc. 134,
84288431 (2012).
27. B. H. Lipshutz, K. Noson, W. Chrisman, A. Lower, J. Am. Chem.
Soc. 125, 87798789 (2003).
28. L. M. Geary, S. K. Woo, J. C. Leung, M. J. Krische, Angew.
Chem. Int. Ed. 51, 29722976 (2012).
29. V. Komanduri, M. J. Krische, J. Am. Chem. Soc. 128,
1644816449 (2006).
30. Y. Sasaki, Y. Horita, C. Zhong, M. Sawamura, H. Ito, Angew.
Chem. Int. Ed. 50, 27782782 (2011).
31. C.-H. Ding, X.-L. Hou, Chem. Rev. 111, 19141937 (2011).
32. S.-L. Shi, L.-W. Xu, K. Oisaki, M. Kanai, M. Shibasaki, J. Am.
Chem. Soc. 132, 66386639 (2010).
33. K. R. Fandrick et al., J. Am. Chem. Soc. 133, 1033210335 (2011).
34. E. Ascic, S. L. Buchwald, J. Am. Chem. Soc. 137, 46664669 (2015).
35. A. M. Whittaker, G. Lalic, Org. Lett. 15, 11121115 (2013).
36. J. Clayden, P. Maclellan, Beilstein J. Org. Chem. 7, 582595 (2011).

forming a surface complex. The ability of ceria

to improve the sinter resistance of metals in automotive catalysts has been explained in terms
of lowering the surface energy of nanoparticles
(enhanced wetting) (13). However, as we show
here, an additional benefit is the ability of ceria
to trap Pt in an atomically dispersed state. Other
catalyst supports have been shown to stabilize Pt
in the form of isolated atomsfor example, pentacoordinated Al3+ sites on alumina (14), FeOx (15),
and the (111) surface of MgAl2O4 (16). However,
when these catalysts are aged at 800C in air
(16, 17), the formation of large crystalline Pt
particles can be detected via x-ray diffraction
(XRD). The complete absence of an XRD peak
for Pt demonstrates that the sample contains
only atomically dispersed species, which can be
confirmed by reactivity measurements, energydispersive spectroscopy (EDS), aberration-corrected
scanning transmission electron microscopy (ACSTEM), and diffuse reflectance infrared Fourier
transform spectroscopy (DRIFTS).
Exposing ceria powders with different surface
facets allowed us to determine the most effective
surface for trapping Pt. Ceria can be synthesized
in well-defined shapes, specifically nanorods,
cubes, and octahedra (18). The similar specific
reactivity of octahedra and nanorods for the
water-gas shift reaction (19) suggests that the
rods expose dominant (111) facets, contrary to
the previously reported morphology of the nanorods (18). Ceria nanoshapes have been studied
for their role in Pt sintering and redispersion (20).
Density functional theory (DFT) calculations by
Bruix et al. (21) suggested that CeO2 (100) surfaces provide the right geometry to stabilize Pt2+
by bonding to O2 in square pockets. Neitzel et al.
(22) showed that the Pt cations are stable on the
surface rather than diffusing into the ceria bulk.
However, the highest temperature achieved in

these previous studies was much lower than the

aging temperature of DOCs (21).
We sought to determine the effectiveness of
various ceria surface facets for trapping Pt at
typical DOC aging conditions, 800C in air. We
introduced ceria powders as a separate phase
by simple physical mixing. Three different ceria
shapes were used: cubes (50 m2/g), nanorods
(90 m2/g), and what we refer to as polyhedral
ceria (89 m2/g). The latter, obtained by calcination of cerium nitrate, consists of nanoparticles
of truncated octahedra, which expose (111) surface facets, because they constitute the most stable low-index surface of ceria (23). The rounded
nature of these polyhedral particles indicates
that other less prominent surfaces, as well as step
edges, must also be exposed (figs. S1 and S2).
The STEM image of the as-prepared Pt/LaAl2O3 catalyst, nominally 1 weight percent (wt %)
(Fig. 1A), shows small Pt particles on the alumina
support. After the sample was aged at 800C for
10 hours in flowing air, large Pt crystallites were
formed (Fig. 1B). After aging, EDS analysis in a
scanning electron microscope (SEM) showed
~0.8 wt % Pt (table S1), which was confirmed via
XRD analysis (24). XRD patterns corroborated
the rapid sintering, as evident from the welldefined, sharp reflections for face-centered cubic
(fcc) Pt in Fig. 1C; the broad reflections and the
background in this pattern come from the alumina support.
The as-prepared 1 wt % Pt/La-Al2O3 catalyst
was physically mixed with ceria powders in a
weight ratio of 2:1 (33 wt % ceria) and then subjected to aging in flowing air. TEM/STEM images
of the physical mixtures after aging show that
the Al2O3 and CeO2 phases in the mixture are
attached to each other (fig. S3). Furthermore, the
STEM image revealed that no Pt was present on
the alumina because it had migrated to the CeO2

phase (figs. S3 and S4) and was trapped, forming

subnanometer Pt species. XRD patterns of the
physically mixed catalysts after aging at 800C
for 10 hours in air are shown in Fig. 1D, curves a
to c. The size of the Pt(111) peak gets smaller in
samples a to c, suggesting improved efficiency for
trapping Pt (decrease in the amount of the crystalline Pt phase). The sample aged for 1 week in
air (Fig. 1D, curve d) shows no Pt peak at all, indicating the complete absence of large Pt particles. Because ceria can scatter x-rays, it causes a
decrease in the size of Pt peak. This effect of ceria
is seen by comparing the XRD pattern of the
aged Pt/La-Al2O3 (Fig. 1C) with the XRD pattern
of the same sample after mixing with ceria (2 parts
Pt/La-Al2O3:1 part ceria) (Fig. 1D, curve e).
The inset in Fig. 1D shows magnified views
of the Pt(111) reflection. The size and sharpness
of this peak are indicative of the crystallite size
of Pt. The splitting of the peak into a doublet,
the K a1 and K a2, is further evidence of large Pt
crystallites. With such large particles, XRD
cannot be used to estimate crystallite dimensions
because size-induced broadening is smaller than
the instrumental broadening. The size of the Pt
peak gives us a visual indication of the amount of
crystalline Pt in these samples. The Rietveld refinement method (fig. S5) was used to quantify
the amount of crystalline Pt visible to XRD, which
is reported for each of these samples in table S2.
The calculated value of 0.81 wt % Pt in the reference sample (Fig. 1D, curve e) agrees with independent EDS analysis of the overall composition
of this sample (table S1). The excellent agreement
between XRD-derived composition and elemental
analysis allows us to estimate the amount of crystalline Pt visible to XRD in each of these samples.
Subtracting the amount of crystalline Pt from the
total Pt gives us the amount of atomically dispersed
Pt in each sample (table S2).

Fig. 1. The transfer of

Pt from alumina to ceria leads to a decrease
in size of the Pt(111)
XRD peak. (A) HAADF
STEM image of 1 wt %
Pt/La-Al2O3 after calcination in air at 350C
for 5 hours. (B) TEM image of this catalyst after
aging at 800C in air for
10 hours. (C) XRD pattern of this catalyst after
aging in air at 800C for
10 hours. The sharp Pt
reflections (vertical arrows) arise from the large
particles seen in the TEM
image (B), whereas the
broad peaks come from
the alumina support.
(D) XRD patterns of physical mixtures of 1 wt % Pt/La-Al2O3 and ceria powders (weight ratio 2:1) after aging at 800C for 10 hours in flowing air [(a) to (c)],
aged for 1 week with polyhedral ceria (d), and a reference sample (e) where the Pt/La-alumina catalyst was first aged as in (C), then mixed with polyhedral ceria
in the same weight ratio.The asterisks show the CeO2 reflections whose peak widths vary as a result of changes in crystallite sizes due to aging.The inset shows
a higher-magnification view of the Pt(111) reflection; the size of the peak corresponds to the amount of crystalline Pt in the sample.


8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 2. Physically mixing ceria with Pt/La-Al2O3 leads to enhanced reactivity. (A) Illustration of Pt nanoparticle sintering, showing how ceria can trap the
mobile Pt to suppress sintering. Cubes appear to be less effective than rods or polyhedral ceria. (B) Light-off curves for CO oxidation on the 1 wt % Pt/La-Al2O3
sample (20 mg of catalyst in the reactor) and the samples physically mixed with ceria (20 mg of 1 wt % Pt/La-Al2O3 mixed with 10 mg of ceria powder) and
then aged at 800C in air. The higher reactivity of the samples containing ceria correlates with the smaller Pt peak in the XRD patterns (inset in Fig. 1).

Fig. 3. Heating in air at 800C

allows mobile Pt species to be
trapped on ceria. (A to D) Representative AC-STEM images of
1 wt % Pt/CeO2-rod [(A) and (B)]
and 1 wt % Pt/CeO2-polyhedra
[(C) and (D)] after aging at 800C
for 10 hours in flowing air. The
darker objects within the ceria
particles are internal voids. The
bright dots on the surface of CeO2
represent mononuclear Pt species
that are stable to treatment in
flowing air at 800C. When the
ceria lattice is aligned with the
electron beam [(B) and (D)], it is
more difficult to distinguish the Pt
because of the strong contrast
from the ceria. The Pt species
reside on the surface of ceria rather
than being embedded within it.

Figure 2A provides a schematic illustration

of how physically mixed ceria traps Pt emitted
from the alumina catalyst, thereby helping to
improve the reactivity for CO oxidation (Fig. 2B).
The smaller the Pt peak in the inset in Fig. 1D,
the more atomically dispersed Pt that is present
on the sample, which leads to higher catalytic
activity. The aged Pt/La-alumina had such large
Pt crystallites that there was negligible CO oxidation reactivity at 225C. Among the nanoshapes,
the physically mixed polyhedral ceria and the
nanorods show similar performance, with the
lowest reactivity seen with the ceria cubes. When
the physically mixed polyhedral ceria was aged
for 1 week with the Pt/La-Al2O3, the reactivity
was comparable to that of the fresh catalyst (per

8 JULY 2016 VOL 353 ISSUE 6295

gram of catalyst), indicating remarkable sinter

resistance (Fig. 2B).
On the basis of the CO oxidation activity shown
in Fig. 2, we calculated the specific reactivity
[turnover frequency (TOF)] for the Pt (table S2
and fig. S6). Chemisorption of H2, O2, and CO
was performed on these catalysts (24) after oxidative pretreatments similar to those used for
the reactivity measurements. No H2 or O2 uptake
was detected; the ratio of chemisorbed CO/Pt
was 0.18. The low CO uptake would suggest a
particle size of ~5 nm, but high-angle annular
dark-field (HAADF) AC-STEM imaging showed
only atomically dispersed Pt species on the polyhedral ceria and on the ceria rod samples (Fig. 3).
The ceria underwent sintering after being heated

at 800C in air, as seen from the peak widths of

the ceria XRD peaks highlighted with asterisks in
Fig. 1D. As a result, the rods and cubes changed
shape and became more rounded (Fig. 3 and
fig. S7). The effect of high-temperature treatment was most severe on the ceria cubes, as seen
from the Ka1-Ka2 splitting of the high-angle ceria
XRD reflections (Fig. 1D, curve a) indicating large
crystallite sizes. We could not image any singleatom Pt species on the ceria cubes; hence, the
concentration of atomically dispersed Pt on the
cubes is apparently much lower (fig. S7).
Because chemisorption of CO was not useful
for counting the number of Pt sites, we used
the amount of atomically dispersed Pt derived
from the XRD measurements (24). The TOF derived in this manner is remarkably similar for
all of these catalysts, within experimental error
(0.12 0.04 s1 at 225C). Surprisingly, the TOF
is also close to that reported for single crystals
of Pt(100) (25) at this temperature (0.12 s1),
even though the atomically dispersed Pt is very
different from a monolithic single crystal in
morphology and nearest neighbors. These TOF
values are lower than those reported on the most
active Pt/CeO2 catalysts (26, 27), but the latter
contain metallic Pt nanoparticles while our catalysts contain atomically dispersed Pt that is
present in ionic form, most likely Pt2+. A reaction
mechanism based on DFT calculations was proposed for single isolated Pt2+ on alumina, and
although the mechanism differs from that on
metallic Pt, the TOF for CO oxidation was comparable to that on 1 wt % Pt/alumina (28).
Because Pt was deposited through the vapor
phase in physically mixed ceria samples, we prepared a second set of catalysts directly on the
same ceria powders using a conventional method, impregnation of chloroplatinic acid. These
1 wt % Pt/ceria samples were then aged at similar SCIENCE

Downloaded from on July 8, 2016


Fig. 4. The atomic dispersion of Pt is preserved during CO oxidation measurements. (A) DRIFTS of the 1 wt % Pt/CeO2 polyhedra during CO oxidation
at 125C after oxidative pretreatment in 10% O2 at 450C. After 30 min of CO oxidation, the CO flow was stopped and spectra recorded at 0, 2, and 10 min while
the oxygen flow continued.There is negligible drop in the symmetrical feature at 2095 cm1 that is assigned to isolated ionic Pt sites on the ceria. (B, C, E, and F)
HAADFAC-STEM images of Pt/ceria rods treated at 800C in flowing air for 10 hours, before [(B) and (C)] and after three cycles of CO oxidation [(E) and (F)] to
300C, showing that the catalyst is stable after reaction and the Pt species remain atomically dispersed. The arrows point to step edges that appear to be
sites where the Pt species are present, rather than the smooth well-defined (111) facets. (D) DRIFTS of a 1 wt % Pt/Al2O3 catalyst after similar pretreatment
as in (A) and after 30 min of CO oxidation at 125C. Spectra were recorded after CO was switched off while the oxygen continued to flow for 0, 5, and 10 min.
The band of CO on metallic Pt (2087 cm1 and 2064 cm1) disappeared rapidly, leaving behind a feature at ~2100 cm1 that corresponds to oxidized Pt.

conditions (800C in air for 10 hours). The XRD

patterns of these aged Pt/ceria samples (fig. S8)
show that a well-defined Pt(111) reflection is seen
only on the Pt/ceria cube sample, but no Pt(111)
reflection is seen on the nanorods or the polyhedral ceria. SEM images (fig. S9) show large Pt
particles on the cube sample, but no such large
particles were detected on the rod or polyhedral ceria samples. The conversion of CO was
marginally higher on the rod samples than on
the polyhedral ceria; the cubes were less reactive
(fig. S10). TEM images of the Pt species in the
physically mixed ceria after aging (fig. S3) are
similar to the Pt/ceria prepared by impregnation
(fig. S11). The method of Pt deposition does not
seem to matter as long as the sample is aged at
elevated temperatures in air. The similarity in
TOF for Pt/alumina and Pt/ceria suggests that
the Pt is not lost into the bulk of the ceria. Solidsolution substitution of platinum for cerium in
bulk ceria is unfavorable, given the smaller ionic
radii of Pt2+ or Pt4+. Platinum cations prefer sixfold octahedral coordination with surrounding
oxygens, in contrast to the eight-fold cube coordination of oxygens surrounding cerium cations
in CeO2.
We examined the Pt/ceria rod samples after
three cycles of CO oxidation going up to 300C.
The Pt was atomically dispersed before (Fig. 4,
B and C) and after reaction (Fig. 4, E and F). To
examine the nature of the Pt during reaction, we
performed DRIFTS of the 1 wt % Pt/polyhedral

CeO2 sample (Fig. 4A) and compared the results

with a 1 wt % Pt/Al2O3 sample (Fig. 4D). Both
samples were subjected to an oxidative treatment (10% O2 in He at 450C) before exposure
to reaction conditions at 125C (CO:O2 = 1.5:1).
A prominent symmetrical band at 2095 cm1
was the only feature seen on the Pt/CeO2 support (Fig. 4A), whereas on Pt/alumina we observed multiple CO adsorption features typical
of metallic Pt nanoparticles, including a band
on ionic Pt at ~2100 cm1 (which is seen most
clearly when the CO is switched off while oxygen continues to flow for 10 min) (Fig. 4D). The
coexistence of Pt ionic species and Pt nanoparticles on alumina and other oxide supports
was also noted by Ding et al. (29). To prepare a
sample that contained exclusively ionic Pt (and
showing a CO adsorption band similar to Fig.
4A), they used a ZSM-5 support and low metal
loading (0.5 wt %). The Pt/ZSM-5 was not stable beyond 1 min during electron microscope
imaging, forming Pt nanoparticles (29). As seen
from the images in Figs. 3 and 4 (24), the Pt/ceria
is stable in the TEM and during repeated CO oxidation tests. The 2095 cm1 band is the only feature seen even at 350C during CO oxidation (fig.
S12), indicating that Pt nanoparticles do not form
under these conditions. As reported by Ding et al.
(29), CO on ionic Pt (2095 cm1) is not reactive at
low temperatures (Fig. 4A), whereas CO on metallic Pt (2064 cm1) reacts readily with flowing
oxygen (Fig. 4D). Despite the transient activity

seen on metallic Pt nanoparticles, both of these

catalysts exhibit very low steady-state activity at
this temperature (Fig. 2 and figs. S13 and S14)
due to strongly bound CO. It is only at higher
temperatures that the Pt nanoparticles on alumina and the single-atom Pt on ceria become
active (Fig. 2) and exhibit similar specific activity
for CO oxidation (table S2).
While ceria helps to trap atomically dispersed
Pt, the strong interaction between Pt and ceria
also helps preserve the surface area of polyhedral
ceria (table S3). We estimate a surface atom
concentration of 1.1 Pt atoms/nm2 in the aged
sample (table S3). A recent study (30) suggests
that step edges on the ceria (111) surface provide
stable sites for Pt2+. By engineering the density
of surface steps, 80% of the Pt loading of 0.18
monolayer (1 monolayer = 7.9 atoms/nm2) could
be stabilized in the form of Pt2+. This corresponds to 1.2 atoms/nm2, which is similar to the
loading of atomically dispersed Pt that we observed. The high step density was achieved by
Dvok et al. (30) through deposition of ceria on
their CeO2 (111) thin films or via heat treatment.
The AC-STEM images, especially those in Fig. 4,
suggest that the Pt occupies step edges (arrowed)
on the ceria rod sample. On the other hand, the
well-defined (111) facet in Fig. 4C does not show
any Pt species. DFT calculations suggest that
the binding energy of Pt to surface steps on
CeO2 (30) exceeds that of Pt on CeO2 (111) and
also the cohesive energy of bulk Pt. The images in
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




1. G. W. Graham et al., Catal. Lett. 116, 18 (2007).

2. M. H. Wiebenga et al., Catal. Today 184, 197204 (2012).
3. T. W. Hansen, A. T. Delariva, S. R. Challa, A. K. Datye,
Acc. Chem. Res. 46, 17201730 (2013).
4. T. R. Johns et al., J. Catal. 328, 151164 (2015).
5. J. A. Kurzman, L. M. Misch, R. Seshadri, Dalton Trans. 42,
1465314667 (2013).
6. C. B. Alcock, G. W. Hooper, Proc. R. Soc. London Ser. A 254,
551561 (1960).
7. G. Cavataio et al., SAE Int. J. Fuels Lubr. 2, 204216 (2009).
8. Y.-F. Yu-Yao, J. T. Kummer, J. Catal. 106, 307312 (1987).
9. J. G. McCarty, K.-H. Lau, D. L. Hildenbrand, Stud. Surf.
Sci. Catal. 111, 601607 (1997).
10. C. Carrillo et al., J. Phys. Chem. Lett. 5, 20892093
11. G. B. McVicker, R. L. Garten, R. T. K. Baker, J. Catal. 54,
129142 (1978).
12. Y. Nagai et al., J. Catal. 242, 103109 (2006).
13. J. A. Farmer, C. T. Campbell, Science 329, 933936 (2010).
14. J. H. Kwak et al., Science 325, 16701673 (2009).
15. B. Qiao et al., Nat. Chem. 3, 634641 (2011).
16. W.-Z. Li et al., Nat. Commun. 4, 2481 (2013).
17. T. R. Johns et al., ChemCatChem 5, 26362645 (2013).
18. H.-X. Mai et al., J. Phys. Chem. B 109, 2438024385
19. S. Agarwal et al., ChemSusChem 6, 18981906 (2013).
20. T. Wu et al., J. Phys. Chem. Lett. 5, 24792483 (2014).
21. A. Bruix et al., Angew. Chem. Int. Ed. 53, 1052510530
22. A. Neitzel et al., J. Phys. Chem. C 120, 98529862 (2016).
23. Z. L. Wang, X. Feng, J. Phys. Chem. B 107, 1356313566
24. See supplementary materials on Science Online.
25. P. J. Berlowitz, C. H. F. Peden, D. W. Goodman, J. Phys. Chem.
92, 52135221 (1988).
26. R. Kopelent et al., Angew. Chem. Int. Ed. 54, 87288731
27. M. Cargnello et al., Science 341, 771773 (2013).
28. M. Moses-DeBusk et al., J. Am. Chem. Soc. 135, 1263412645
29. K. Ding et al., Science 350, 189192 (2015).
30. F. Dvok et al., Nat. Commun. 7, 10801 (2016).

recording the AC-STEM images and D. Kunwar for assistance

in catalyst preparation.
Materials and Methods

Supported by NSF GOALI grant CBET-1438765 (J.J., H.X., S.R.C.,

A.K.D.), General Motors Global R&D (G.Q., S.O., and M.H.W.),
U.S. Department of Energy grant DE-FG02-05ER15712 (A.T.D., E.J.P.,
A.K.D., X.I.P.H., and Y.W.), and the Center for Biorenewable Chemicals
funded by NSF grant EEC-0813570 (H.X., H.P., and
A.K.D.). This work made use of the JEOL JEM-ARM200CF at the
University of Illinois at Chicago. We thank A. Nicholls for


8 JULY 2016 VOL 353 ISSUE 6295

26 April 2016; accepted 13 June 2016



Tail use improves performance on

soft substrates in models of early
vertebrate land locomotors
Benjamin McInroe,1* Henry C. Astley,1* Chaohui Gong,2 Sandy M. Kawano,3
Perrin E. Schiebel,1 Jennifer M. Rieser,1 Howie Choset,2
Richard W. Blob,4 Daniel I. Goldman1,5
In the evolutionary transition from an aquatic to a terrestrial environment, early tetrapods
faced the challenges of terrestrial locomotion on flowable substrates, such as sand and
mud of variable stiffness and incline. The morphology and range of motion of appendages
can be revealed in fossils; however, biological and robophysical studies of modern taxa
have shown that movement on such substrates can be sensitive to small changes in
appendage use. Using a biological model (the mudskipper), a physical robot model,
granular drag measurements, and theoretical tools from geometric mechanics, we
demonstrate how tail use can improve robustness to variable limb use and substrate
conditions. We hypothesize that properly coordinated tail movements could have provided
a substantial benefit for the earliest vertebrates to move on land.

uring the vertebrate invasion of land, 385

to 360 million years ago, early tetrapods
and relatives faced a variety of challenges
(1), including locomotion in terrestrial environments. Terrestrial locomotion relies
on interactions between the body and substrate
to generate propulsive forces, but the interaction
between the organism and some substrates may
be complex. Fossil evidence indicates that tetrapods emerged from water in near-shore habitats,
where they likely encountered flowable soft substrates such as sands and muds (2, 3). These
substrates exhibit properties of solids and fluids,
either jamming or yielding (plastic deformation
of the material) depending on how they are
loaded (4) and sloped (5).
The challenge of movement on flowable substrates therefore arises from the complexity of
interactions between the substrate and the organism. Even on level deformable substrates,
subtle variations in limb morphology (6) and
kinematics (7) can lead to substantial differences
in performance. Furthermore, interactions between appendages and these substrates leave


Figs. S1 to S14
Tables S1 to S3
References (3135)

School of Physics, Georgia Institute of Technology, Atlanta,

GA, USA. 2Robotics Institute, Carnegie Mellon University,
Pittsburgh, PA, USA. 3National Institute for Mathematical and
Biological Synthesis, University of Tennessee, Knoxville, TN,
USA. 4Department of Biological Sciences, Clemson University,
Clemson, SC, USA. 5School of Biology, Georgia Institute of
Technology, Atlanta, GA, USA.
*These authors contributed equally to this work. Corresponding
author. Email:

local disturbances, which can influence subsequent interactions, sometimes leading to deteriorating locomotor performance and eventual
total locomotor failure (8). As substrate slope increases, yield forces decrease and downhill material flow becomes important, reducing the range
of effective locomotor strategies (5).
The use of an additional locomotor structure
that can be independently coordinated may allow
a greater range of effective behaviors, even in the
absence of derived limb morphology and sophisticated motor patterns. We propose that the tail
could have been a critical locomotor structure for
early tetrapods. In addition to being a primary
driver of aquatic locomotion, tails play major
roles in the propulsion of many modern fishes
during terrestrial locomotion (912) and can be
used as inertial reorientation appendages in some
tetrapods (13, 14). Thus, the use of a prominent
tail [as seen in fossil taxa (1517) (Fig. 1A)] may
have increased locomotor robustness to environmental and kinematic variables.
Evaluating locomotor performance for extinct
taxa is challenging (18, 19), in part because the
sensitivity of locomotion on complex substrates
to kinematics and control strategies cannot necessarily be inferred from range of motion and
morphology (7). Therefore, to test our hypothesis,
we used three complementary modeling methods
(Fig. 1): a model organism, a robophysical model,
and a mathematical model. We made several choices
governing our modeling approaches. In our locomotors, we modeled symmetrical, forelimb-driven SCIENCE

Downloaded from on July 8, 2016

Figs. 3 and 4 show Pt species located randomly

on the ceria surfaces (not embedded in the ceria),
with no preference for specific facets.
Atom trapping should be broadly applicable
as a method for preparing single-atom catalysts.
The approach requires a supply of mobile atoms
and a support that can bind the mobile species.
Conditions that are conducive to Ostwald ripening,
which normally is implicated in the degradation
of catalysts (3), are ideal because mobile species
are continually being generated. In our work, at
the aging temperature of 800C in air, mobile
PtO2 is rapidly emitted; the estimated lifetime is
only a few seconds for a 5-nm Pt crystallite (24).
Surface species such as hydroxyls and carbonates,
which could prevent the trapping of mobile species, would have desorbed at high temperatures,
providing a clean surface for the formation of
covalent metal oxide bonds that are needed to
stabilize single atoms. Trapping of atoms provides a plausible explanation for the role of ceria
in slowing the rates of Ostwald ripening and may
help to explain how other supports modify the
rates of catalyst sintering.


= 20

t = 0 ms
t = 0 ms

10 cm

250 ms

x (BL)

y (cm)

4 cm

2 (deg)

= 0

210 ms



1 cm

4 cm

Fig. 1. Target and model systems for understanding early tetrapod locomotion on granular media.
(A) A reconstruction of Ichthyostega (360 million
years ago), an example of an early tetrapod body
plan, by Raul Martin. (B) Skeletal reconstruction of
Ichthyostega, an example of an early tetrapod body
plan [from (20)], highlighting the pectoral limbs (green)
and tail (blue). (C) The mudskipper (Periophthalmus
barbarus), a biological model for early terrestrial locomotors. (D) A microcomputed tomography scan
reconstruction of a mudskipper skeleton, highlighting the pectoral fin (green) and tail (blue). (E) The
MuddyBot, a 3D printed robot developed to model
the locomotion of crutching early tetrapods. Limbs
are in green and the tail is in blue.

crutching locomotion (rather than salamanderlike movement) in accordance with recent studies
of Ichthyostega (20, 21); this choice enabled simplicity of control (coordinating two appendages
rather than four) and obviated the need for continuous, stable support of an elevated body. We note
that we are not strictly modeling Ichthyostega,
nor any specific fossil taxon associated with the
water-land transition, but are instead seeking
general principles underlying limbed, crutching
locomotion on yielding media.
Our biological model is the mudskipper (Periophthalmus barbarus) (12, 18, 2024), a small fish that
frequently moves terrestrially using synchronous motions of the pectoral fins, although these
animals can use their tails for rapid jumping

620 ms

780 ms

of Tail Use

570 ms

1 cm





Fig. 2. Mudskipper locomotion on granular media at different substrate inclines (q). (A and B) Dorsalview video frames of a mudskipper fish on dry, loose sand inclined at 0 (A) and 20 (B) (movies S1 and
S2). Yellow solid lines along the longest tail fin ray and from between the eyes to the anterior edge of the
dorsal fin are used to compute the tail angle (a2) in (E). The tail is not used propulsively in (A), although it
moves slightly; in (B), the tail is used for propulsion. (C to E) Horizontal forward displacement per cycle (Dx)
in body lengths (BL) for a single trial (C), vertical displacement (Dy, measured from eye) (D), and tail angle
(a2) (E) of mudskippers on sand at 20 incline. Cycles without tail use are indicated by green regions; cycles
with tail use are indicated by blue regions (determined from video inspection). Missing values of a2 are
when the tail fin was out of view. (F) Dx at all inclines (q) for steps without (green) and with (blue) tail use.
Error bars denote SD. (G) Percentage of cycles with propulsive tail use across substrate inclines.

despite unspecialized tail morphology (25).

Our robophysical model (Fig. 1E) (26) was designed with morphology representing the
simplest possible version of a crutching locomotor,
allowing us to systematically explore performance over a range of locomotor movements,
including those movements not used by our biological model. Our mathematical model relies on
the framework of geometric mechanics (27, 28)
and allows us to understand how certain aspects
of performance relate to coordination of limbs
and tail. Our use of dry granular media was
another modeling choice: The first vertebrates to
move on land likely did so in wet shoreline habitats (such as mudflats) with properties that differ
from those of dry granular media, but these
media are united in displaying plastic deformation once the yield stress is exceeded (2931). Although partially wet soils can display cohesion
that increases yield stresses and results in larger
memory effects, these rheological similarities
[combined with the difficulty of preparing large
volumes of standardized wet granular media (30)]
led us to use dry media to model flowable substrates in our trials (30, 31).
Mudskippers (N = 6) were capable of effective
locomotion over a level (q = 0) granular substrate [loose-packed dry oolite sand (table S1)]

using a crutching gait actuated by synchronous

retraction of the pectoral fins with the tail angle
(a2) almost straight (21 22; all measurements
are means SD) (Fig. 2A and movie S1), moving an
average forward distance per cycle (Dx) of 0.21
0.03 body lengths (BL) (Fig. 2F). During locomotion, mudskippers were able to fold the fin rays
of the caudal and anal fins away from the substrate, allowing terrestrial locomotion without
damage to or interference from these structures
(Fig. 2B).
In some locomotor cycles, mudskippers used
their tails (Fig. 2B and movie S2), bending the
tail and planting the distal end of the tail and
fin rays in the sand approximately orthogonal
to the body axis (119 21) (Fig. 2E) before
retracting the pectoral fins and straightening the
tail during the propulsive phase (Fig. 2B). Tracks
left by these trials typically consisted of a series of
paired impressions from the pectoral fins (sometimes obscured by yielding flow from subsequent steps, especially on inclines), with tail use
leaving an additional impression offset to the
right or left and overlapping the previous fin
impressions (Fig. 2, A and B, and movies S1 and
S2). Steps using the tail resulted in a higher
displacement of 0.28 0.08 BL on horizontal
media (Fig. 2, C, D, and F).
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016

370 ms


No Tail

x (BL)

380 ms



Time (sec)


10 cm


8 JULY 2016 VOL 353 ISSUE 6295

In each trial, the robot performed a total of six

limb cycles, the maximum number possible for
the size of the bed. Insufficient first-step displacement has been shown to produce interactions
with the previously disturbed media, resulting
in lower displacement and further interactions
until complete failure (stranding) (8). Conversely,
if the first step is sufficient to prevent this interaction, the likelihood of failure decreases. In our
experiments, few configurations produced intermediate displacements that led to decaying performance; most were either consistently successful
or immediate failures (fig. S1). These trials were
not sufficiently numerous to determine whether
tail use altered the rate of decay, although it may
do so indirectly if it increases first-step displacement. Consequently, we present data for only the
first step (Dx1).
To characterize how the adduction, supination,
and tail use parameters affected the robots locomotion in undisturbed media, we measured Dx1
for three trials per configuration (Fig. 3, D and E).
During these trials, high values of adduction
(which resulted in lifting much of the body clear
of the substrate) and use of the tail improved firststep displacement, although not in a simple, addi-

tive manner; supination angle had an additional,

minor influence (Fig. 3, D and E). On the level
substrate, increasing adduction angle led to the
largest performance increase, with supination
angles of 30 and tail use resulting in modest
improvements at lower adduction angles but offering minimal improvement over performance at
the highest adduction angle (Fig. 3, B and D).
At higher substrate angles with consequently
lower granular yield forces, the role of the tail
became dominant; effective locomotion without
the tail was possible only at the highest adduction and lowest supination angles, but the use
of the tail allowed locomotion over a wide range
of limb kinematics (Fig. 3, C and E). In many of
these cases, the use of the tail was the difference
between success and failure [at q = 20, there
were 15 failures among the 30 different configurations of f and y (Fig. 3, D and E)]. Thus, the
tail was not simply a uniform addition of propulsive force conferring a uniform advantage, as it did
little to improve performance under near-optimal
conditions. Instead, the tail had the greatest benefit when locomotion was otherwise compromised
or ineffective as a result of low adduction angle
or high substrate angle. A second set of trials

= 0



x (BL)

= 20

Tail Use

No Tail

= 0

2 3 4 5 6
Step Number

= 20





x1 (BL)






x1 (BL)

As substrate incline angle increased, crutching

with only the pectoral fins became less effective
(Fig. 2F) and the frequency of steps for which the
tail was used propulsively increased, from 6% on
level substrate to 36% on substrate inclined at 10
and to 55% on substrate inclined at 20 (Fig. 2G).
In addition to increasing displacement per cycle,
tail use prevented downhill slip on inclined
media when the tail was planted. The disparity
in forward displacement increased between steps
with and without tail use as substrate angle increased (Fig. 2F). At a 10 incline, mudskippers
moved forward an average of 0.16 0.03 BL with
each step, versus 0.18 0.08 BL when the tail was
used (Fig. 2F). When the substrate was inclined
to 20, displacement was 0.07 0.03 BL without
tail use, increasing to 0.14 0.07 BL with tail use
(Fig. 2F). Further, mudskippers would occasionally
jump clear of the test arena via tail-powered movements, indicating that the use of tails during
crutching locomotion is controlled and submaximal.
A robophysical model of a crutching locomotor with limbs (Fig. 3A) allowed systematic
testing of how locomotor performance in realistic granular environments was affected by variations in foot placement, limb adduction, and
tail use (7). We elected to use a highly simplified
morphology (a pair of laterally positioned, synchronously moving forelimbs and a posterior
tail) (Fig. 3A) in order to focus on overarching
locomotor control principles (templates) as opposed to the details of their anatomical implementation (anchors) (32). Although this model
is simplified, it possesses some of the degrees of
freedom seen in the mudskippers, namely the
ability to control both body lifting (via limb adduction) and the interface with the substrate
(limb supination) as well as use of the tail.
In addition to varying the incline (q = 0, 10,
and 20) of the granular material, we varied three
parameters of the robot: limb adduction angle
(y), limb supination angle (f), and presence or
absence of tail use (Fig. 3). Adduction angle and
the resultant lifting of the body is known to affect
locomotor performance in similar robots on granular media (8) and may have been critical to
terrestrial locomotion in early tetrapods (20). Supination angle, a simplified model of differences
in limb placement, varied from a vertical limb insertion into the media (0) to nearly flat (60) in
15 increments. The tail was either used or not
used, and was lifted clear of the media when not
in use (resulting in the posterior portion of the
robot dragging in the media). In all cases, tail use
induced alternating lateral rotations and displacements; although these yaw movements were often
small (<10), they explain the few instances in
which tail use was detrimental. To prevent damage to the motors, instead of sand we used two
granular materials, loosely packed poppy seeds
and spherical plastic particles (table S1) (33); prior
work has shown that these substrates function
well as models of more natural granular media
(8, 33). Although the robots movements were
slower than those of the mudskippers, all three substrates showed no dependence of force on speed
over the ranges observed (5 to 10 cm/s) (fig. S6).




Fig. 3. Robophysical experiments on granular media at various substrate inclines (q). (A) 3D model
of MuddyBot, showing ranges of motion for limb retraction (green arrow, 60) and tail motion (blue
arrow, 90). Limb adduction (y, 5 to 20, where 0 is horizontal) and limb supination (f, 0 to 60,
where 0 is vertical to the limb) are labeled, with other arrows showing directions of limb and tail motion
during thrust phase. (B and C) Kinematics of a single trial of MuddyBot (f = 15, y = 15) moving for six
cycles, without tail use (green) and with tail use (blue), on level (q = 0) (B) and inclined (q = 20) (C)
poppy seeds. (D and E) First-step net displacement versus adduction and supination angles on q = 0
(D) and q = 20 (E) poppy seeds. Blue shading shows regions of identical supination angle for clarity.
Vertical lines denote SD > 0.01. Gray shading indicates negative values. SCIENCE

Downloaded from on July 8, 2016


mechanics. This method, developed as a theoretical framework to elucidate principles of movement (27), describes how the self-deformation of
the body (in our case, limb or tail movements)
(Fig. 4A) generates net translation (or rotation)
of the body. Geometric mechanics has been useful for understanding robot swimming (cyclic
self-deformation due to traveling-wave body
undulation) within granular media in which






F /F


conducted using a different granular material

showed qualitatively similar results [spherical
plastic particles (table S1)], indicating that these
results are robust to different granular media
of different particle size and friction (figs. S2
and S5).
To gain insight into effective coordination of
locomotor structures, we created a mathematical planar model of the robot using geometric







Drag Angle, ()



















Displacement (BL/Cycle)






= 0

= 20














1 (rad)


1 (rad)

Fig. 4. Geometric mechanics model of MuddyBot locomotion. (A) A diagram of the simulation,
showing the limb angle (a1) and tail angle (a2) as well as the reaction forces used to compute the local
connection (see supplementary materials) between body deformation and body displacement. (B) The ratio
of forces parallel and perpendicular to the limb surface during poppy seed drag experiments at various drag
angles to the direction of motion, insertion depths, and substrate slopes. Intrusion depths are 1 cm (crosses)
and 3 cm (circles), with q = 0 (brown), 20 uphill (orange), and 20 downhill (green).The black line represents
the ratio of perpendicular and parallel equations (fitted independently; see supplementary materials).
(C) First-step displacement (without slip) of the robot and simulation at all inclines (y = 20, f = 0),
showing close agreement between the simulation and robot performance. Red and pink bars indicate the
optimal and worst gaits, respectively. (D and E) Local connection vector field for q = 0 (D) and q = 20 (E)
media, showing the limb only with tail dragging (purple) and tail gaits (light blue), as well as calculated
optimal (red) and worst (pink) gaits.


inertial forces were small relative to frictional

forces (28).
The geometric mechanics framework relies on
construction of a local connection (34), which
can be visualized as vector fields representing
the link between small self-deformations (changes
in the locomotors shape space, the set of internal
shapes the mechanism can assume, here defined
by planar limb and tail angles; Fig. 4A) and the
resulting movement in world space (Fig. 4, D and
E). For any given body configuration (a point in
the diagrams in Fig. 4, D and E), the corresponding vector indicates the body deformation
pattern that results in maximal forward movement. That is, if the robot begins from an arbitrary limb and tail angle combination (Fig. 4A),
which corresponds to a horizontal and vertical
location on the vector field (Fig. 4, D and E), a selfdeformation parallel to the vector at that location
will produce the greatest world-space incremental
displacement, whereas self-deformations perpendicular to that vector will produce zero displacement. The overall pattern of a vector field allows
visualization of how a time-varying pattern of selfdeformationsrepresented as a path in the shape
spaceresults in translation (or rotation). The net
movement for a given sequence of limb and tail
movements can be evaluated via a line integral
in the vector field, with more effective motor patterns represented as paths that locally align more
closely with vectors and pass through vectors of
larger magnitudes (Fig. 4, D and E) (24).
Construction of local connection vector fields
requires knowledge of how limb and tail segments
experience drag forces. Because fundamental
equations of motion for granular drag in the regime relevant to our robot studies do not exist, we
generated the vector fields in Fig. 4, D and E, by
empirically estimating the forces acting on a
robot limb moving through granular media in
plate drag experiments (figs. S3, S5, and S6)
[assuming the material was continuously deforming and thus in the frictional fluid regime
(33, 35)]. Similar to previous studies (5, 33, 35),
the measured force was a function of the drag
angle between the limb tangent and velocity
vectors (Fig. 4B, inset), depth of intrusion into
the media, and media incline, and was insensitive to speed (within the relevant range).
In such non-inertial limb- and tail-driven locomotion, the interaction between the granular media and the limb is governed by the ratio of the
perpendicular thrust forces to parallel drag forces.
These ratios collapsed to a single curve across
inclines, depths (Fig. 4B), and granular media
(fig. S5 and supplementary materials). This collapse suggests that the change in performance on
granular slopes in the crutching locomotion may
be a consequence of the effect of gravity on the
body. Because our drag measurements were made
in freshly prepared media, and because we could
not model the possible interactions with the
footprints of previous steps (8), we confined our
analysis to the first step, as with the robot. Further,
only high adduction angles were considered, so
that we could avoid the effects of an accumulating
pile of granular media at the front of the robot
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




1. J. A. Clack, Gaining Ground: The Origin and Evolution of

Tetrapods (Indiana Univ. Press, 2002).
2. E. B. Daeschler, N. H. Shubin, F. A. Jenkins Jr., Nature 440,
757763 (2006).
3. A. Blieck et al., Geol. Soc. London Spec. Publ. 278, 219235
4. N. Gravish, P. B. Umbanhowar, D. I. Goldman, Phys. Rev. Lett.
105, 128301 (2010).


8 JULY 2016 VOL 353 ISSUE 6295

5. H. Marvi et al., Science 346, 224229 (2014).

6. F. Qian et al., Bioinspir. Biomim. 10, 056014 (2015).
7. C. Li, P. B. Umbanhowar, H. Komsuoglu, D. E. Koditschek,
D. I. Goldman, Proc. Natl. Acad. Sci. U.S.A. 106, 30293034
8. N. Mazouchova, P. B. Umbanhowar, D. I. Goldman,
Bioinspir. Biomim. 8, 026007 (2013).
9. J. Davenport, A. K. M. A. Matin, J. Fish Biol. 37, 175184 (1990).
10. A. G. Johnels, Oikos 8, 122 (1957).
11. C. M. Pace, A. C. Gibb, J. Exp. Biol. 214, 530537 (2011).
12. C. M. Pace, A. C. Gibb, J. Fish Biol. 84, 639660 (2014).
13. A. Jusufi, D. I. Goldman, S. Revzen, R. J. Full, Proc. Natl. Acad.
Sci. U.S.A. 105, 42154219 (2008).
14. T. Libby et al., Nature 481, 181184 (2012).
15. E. Jarvik, Medd. Gronl. 114, 190 (1952).
16. E. Jarvik, The Devonian Tetrapod Ichthyostega (Indiana Univ.
Press, 1996).
17. M. I. Coates, Trans. R. Soc. Edinb. Earth Sci. 87, 363421 (1996).
18. R. W. Blob, Paleobiology 27, 1438 (2001).
19. M. A. Ashley-Ross, S. T. Hsieh, A. C. Gibb, R. W. Blob,
Integr. Comp. Biol. 53, 192196 (2013).
20. S. E. Pierce, J. A. Clack, J. R. Hutchinson, Nature 486, 523526 (2012).
21. S. E. Pierce, J. R. Hutchinson, J. A. Clack, Integr. Comp. Biol.
53, 209223 (2013).
22. E. M. Standen, T. Y. Du, H. C. E. Larsson, Nature 513, 5458 (2014).
23. S. M. Kawano, R. W. Blob, Integr. Comp. Biol. 53, 283294 (2013).
24. V. A. Harris, Proc. Zool. Soc. London 134, 107135 (1960).
25. B. O. Swanson, A. C. Gibb, J. Exp. Biol. 207, 40374044 (2004).
26. J. Aguilar, D. I. Goldman, Nat. Phys. 12, 278 (2016).
27. A. Shapere, F. Wilczek, Phys. Rev. Lett. 58, 20512054 (1987).
28. R. L. Hatton, Y. Ding, H. Choset, D. I. Goldman, Phys. Rev. Lett.
110, 078101 (2013).
29. N. Mitarai, F. Nori, Adv. Phys. 55, 145 (2006).
30. S. S. Sharpe, R. Kuckuk, D. I. Goldman, Phys. Biol. 12, 046009 (2015).
31. H. Askari, K. Kamrin,

32. K. Nishikawa et al., Integr. Comp. Biol. 47, 1654 (2007).

33. C. Li, T. Zhang, D. I. Goldman, Science 339, 14081412
34. S. D. Kelly, R. M. Murray, J. Robot. Syst. 12, 417431 (1995).
35. R. D. Maladen, Y. Ding, C. Li, D. I. Goldman, Science 325,
314318 (2009).
36. S. J. Gould, E. S. Vrba, Paleobiology 8, 415 (1982).
37. J. Clack, Palaeogeogr. Palaeoclimatol. Palaeoecol. 130, 227250
38. S. Curth, M. S. Fischer, J. A. Nyakatura, Ichnos 21, 3243 (2014).

Supported by NSF grants PoLS PHY-1205878, PHY-1150760,

and CMMI-1361778, Army Research Office (ARO) grant W911NF-11-10514, and the ARL MAST CTA (D.I.G.); ARO Robotics CTA and NSF
National Robotics Initiative IIS-1426655 (H.C.); NSF grants IOS0517340 and IOS-0817794 (R.W.B.); GT UROP and the GT PURA Travel
Grant (B.M.); a Clemson University Wade Stackhouse Fellowship,
NSF award DBI-1300426, and the University of Tennessee, Knoxville
(S.M.K.); and the U.S. Department of Defense, Air Force Office of
Scientific Research, National Defense Science and Engineering
Graduate (NDSEG) Fellowship, 32 CFR 168a (P.E.S.). The authors
declare no conflicts of interest. Data are available from the
corresponding author upon request.
Materials and Methods
Supplementary Text
Figs. S1 to S6
Table S1
Movies S1 to S6
References (39, 40)
16 December 2015; accepted 26 May 2016


Phototactic guidance of a
tissue-engineered soft-robotic ray
Sung-Jin Park,1 Mattia Gazzola,2* Kyung Soo Park,3,4 Shirley Park,5
Valentina Di Santo,6 Erin L. Blevins,6 Johan U. Lind,1 Patrick H. Campbell,1
Stephanie Dauth,1 Andrew K. Capulli,1 Francesco S. Pasqualini,1 Seungkuk Ahn,1
Alexander Cho,1 Hongyan Yuan,1|| Ben M. Maoz,1 Ragu Vijaykumar,5
Jeong-Woo Choi,3,4 Karl Deisseroth,5,7 George V. Lauder,6
L. Mahadevan,2,8 Kevin Kit Parker1,4
Inspired by the relatively simple morphological blueprint provided by batoid fish such as
stingrays and skates, we created a biohybrid system that enables an artificial animala
tissue-engineered rayto swim and phototactically follow a light cue. By patterning
dissociated rat cardiomyocytes on an elastomeric body enclosing a microfabricated gold
skeleton, we replicated fish morphology at 1 =10 scale and captured basic fin deflection
patterns of batoid fish. Optogenetics allows for phototactic guidance, steering, and turning
maneuvers. Optical stimulation induced sequential muscle activation via serpentinepatterned muscle circuits, leading to coordinated undulatory swimming. The speed and
direction of the ray was controlled by modulating light frequency and by independently
eliciting right and left fins, allowing the biohybrid machine to maneuver through an
obstacle course.

ioinspired design, as applied to robotics,

aims at implementing naturally occurring
features such as soft materials, morphologies, gaits, and control mechanisms in
artificial settings in order to improve performance (14). For example, recent soft-robotics
studies raised awareness on the importance of

material properties (3, 4), shifting the focus from

rigid elements to soft materials, whereas other
investigations report successful mimicry of gaits
or morphological features inspired by insects (5, 6),
fish (7, 8), snakes (9), salamanders (10), and cheetahs
(11). Although recent advances have the promise
of bridging the performance gap with animals, SCIENCE

Downloaded from on July 8, 2016

when the body was insufficiently lifted. Additionally, when the tail was lifted clear of the media
in robot experiments, the posterior motor and
mounting structures intruded into and dragged
through the media, resulting in an intrusion that
was difficult to model; therefore, we simulated
trials with the tail intruding into the media in
the same configuration as at the end of tailthrusting behavior, which yielded similar performance. To test these assumptions, we compared
results to a subset of robot trials, and obtained
good agreement between simulation and experiment (Fig. 4C and fig. S4).
The change in the patterns of the local connection vector fields revealed how limbs and tail could
coordinate to produce movement (Fig. 4, C to E).
For example, these fields demonstrated that the
tail was not uniformly beneficial in all situations,
nor even substantially beneficial in horizontal
movement, in which the vertical component of the
vectors (tail contribution) was small. However,
as surface incline angle increased, the horizontal magnitudes of connection vectors decreased,
indicating reduced efficacy of limb-only taildragging gaits (a horizontal path across the vector field). The relatively larger vertical component
across more of the shape space indicated the increased importance of the tail to forward movement. The optimal gait for both inclines was close
to the synchronous thrusting used by the robot
and mudskipper, yielding similar displacements
(Fig. 4, C to E); phase lag between initiation of
limb and tail movement was suboptimal and, in
one case per incline, yielded the worst possible
gait (Fig. 4, D and E). Improper use of the tail
resulted in substantially lower performance than
simply allowing it to drag (Fig. 4, C to E). Additionally, the generally downward direction of
the vectors in both fields demonstrate that purely
tail-powered locomotion (a vertical path down
the right of the vector field) can produce forward
motion, as seen in some extant fish (12).
Our results from a biological analog of early
tetrapods and robophysical and mathematical
models demonstrate that the tail can play an
important role in limb-driven crutching locomotion on inclined granular substrates by making
locomotors more robust to suboptimal kinematics
and substrate conditions. This suggests that the
sizable, well-ossified (and presumably well-muscled)
tails of early tetrapods (1517), originally used for
swimming, may have been co-opted to promote
reliable locomotion over challenging substrates,
providing an exaptation (36) that facilitated their
invasion of land. Although evidence of tail use is
absent among the few fossil trackways attributed
to early tetrapods (37, 38), tail use might be evident in trackways formed on inclined shores.

the current soft-robotic actuators based on, for

instance, electroactive polymers, shape memory
alloys, or pressurized fluids are yet to mature to
the point of replicating the high-resolution complex movements of biological muscles (3, 4).
In this context, biosensors and bioactuators
(12) are intriguing alternatives because they can
intrinsically respond to a number of control inputs
(such as electric fields and optical stimulation).
Thanks to recent advances in genetic tools (13)
and tissue engineering (12), these responses can
be altered and tuned across a wide range of time
and length scales. Some pioneering studies have
exploited these technologies for self-propulsion,
developing miniaturized walking machines (1416)
and flagellar (17) or jellyfish-inspired (18) swimming devices. These biohybrid systems operate at
high energy efficiency and harvest power from
energy-dense, locally available nutrients, although
at present they require specialized environments
(physiological solutions) that may limit their applicability. Moreover, these biohybrid locomotors
lack the reflexive control (9, 19) necessary to enable
adaptive maneuvering and thus the ability to respond to spatiotemporally varying external stimuli.
We designed, built, and tested a tissue-engineered
analog of a batoid fish such as stingrays and
skates. By combining soft materials and tissue
engineering with optogenetics, we created an integrated sensory-motor system that allowed for
coordinated undulating fin movement and phototactically controlled locomotion that is guided via
light stimuli. We drew from fish morphology, neuromuscular dynamics, and gait control to implement a living, biohybrid system that leads to
robust and reproducible locomotion and turning
maneuvers. Batoid fish are ideal biological models
in robotics (8) because their nearly planar bauplan is characterized by a broad dorsoventral disk,
with a flattened body and extended pectoral fins,
that enhances stability against roll (20). They swim
with high energy efficiency (21) by generating
spanwise bending deformations and chordwise

front-to-rear undulatory motion (Fig. 1A and movie

S1) (20, 22) via the sequential activation of pectoral
fin muscles. This undulatory gait allows slender
aquatic animals to channel body movement into
forward motion by exchanging momentum with
the fluid (7, 22) and is a convergent mode of aquatic
propulsion (23). Moreover, batoids can use these
undulatory modes for fine maneuvering and turning by independently and asymmetrically actuating their pectoral fins (24). Inspired by batoids, we
reverse-engineered their musculoskeletal structure (Fig. 1B and fig. S1) via a four-layered architecture (Fig. 1C): a three-dimensional elastomer
[polydimethylsiloxane (PDMS)] body, cast via a
titanium mold (fig. S2A and movie S2); a chemically neutral skeleton fabricated by means of
thermal evaporation of gold through a custom
designed shadow mask; a thin interstitial elastomer layer obtained by spin-coating; and last, a
layer of aligned rat cardiomyocytes generated via
microcontact printing of fibronectin (fig. S3) (25).
This design yielded a tissue-engineered ray
with a single muscle layer capable of downward
contraction (Fig. 1, C and D). Upward contraction
would require a second layer of muscle that acts
antagonistically to the upper layer. To minimize
the complexity of our design, we instead used an
asymmetrical stiff gold skeleton that stores elastic
energy during the downstroke and rebounds during the subsequent relaxation phase. Inspired by
histological analysis (fig. S1) as well as theoretical
considerations (26), we channeled muscular work
and elastic energy into forward motion by breaking fore-aft symmetry through a varying body
rigidity along the anterior-posterior axis. This was
achieved via a thicker body and a denser and

radially further reaching skeleton pattern in the

front (Figs. 1, C and D, and 2A and fig. S2A). Likewise, along the proximal-distal axis, flexibility
of the fins was enhanced by gradually reducing
their thickness (fig. S2A). Last, the PDMS mixture
was adjusted (25) to provide body rigidity and fin
flexibility while conserving overall neutral buoyancy (fig. S2).
The composite supporting structure described
above was coupled to a tissue-engineered muscle
layer, which captured the salient musculoskeletal
features of batoid fish (Fig. 2B). At the mesoscale
(50 mm to 5 mm), batoid myofibers are tightly
bundled and are aligned in the radial direction
parallel to the rays of the skeleton (Fig. 2B). At
the microscale (1 to 10 mm), the Z-lines of the
sarcomeres (the cell forcegenerating units) (Fig.
2C) present strong nematic alignment perpendicular to individual myofibers so as to focus contraction forces. Following this layout, the muscles
and sarcomeres of the tissue-engineered ray were
designed to orient radially from the body and
parallel to the gold skeleton rays (Fig. 2B and
fig. S4), and the Z-lines of the sarcomeres were
engineered to be perpendicular to the skeleton
rays through microscale patterning of fibronectin
(Fig. 2C and fig. S4).
Last, to mimic the sensory-somatic nervous
system that controls the sequential activation of
fin muscles in the batoid fish, we recast the neurocontrol problem as a design problem. A possible
solution is given by serpentine-patterned muscle
circuits that physically determine the propagation of muscle contraction in space and time
(Figs. 1D and 2, D and E), leading to hardwired
coordination that could be triggered with external

Disease Biophysics Group, Wyss Institute for Biologically

Inspired Engineering, John A. Paulson School of Engineering and
Applied Sciences, Harvard University, Cambridge, MA 02138,
USA. 2John A. Paulson School of Engineering and Applied
Sciences, Harvard University, Cambridge, MA 02138, USA.
Department of Chemical and Biomolecular Engineering, Sogang
University, Seoul 121-742, Korea. 4Sogang-Harvard Research
Center for Disease Biophysics, Sogang University, Seoul 121742, Korea. 5Department of Bioengineering, Stanford University,
Stanford, CA 94305, USA. 6Museum of Comparative Zoology,
Harvard University, Cambridge, MA 02138, USA. 7Department of
Psychiatry and Behavioral Sciences and the Howard Hughes
Medical Institute, Stanford University, Stanford, CA 94305, USA.
Department of Organismic and Evolutionary Biology,
Department of Physics, Wyss Institute for Biologically Inspired
Engineering, Kavli Institute for Nanobio Science and Technology,
Harvard University, Cambridge, MA 02138S, USA.
*Present address: Department of Mechanical Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, IL
61801, USA. Present address: Department of Biomedical Engineering, University of Michigan, Ann Arbor, MI 48109, USA.
Present address: Department of Cardiovascular Medicine, Stanford
University Medical Center, Stanford, CA 94305, USA. Present
address: The Winsor School, Boston, MA 02215, USA. ||Present
address: Department of Mechanical, Industrial and Systems
Engineering, University of Rhode Island, Kingston, RI 02881, USA.
Corresponding author. Email:


Fig. 1. Bioinspired concept design of the tissue-engineered ray. (A) A live Little skate, Leucoraja
erinacea, swimming and (B) its musculoskeletal structure. (C to E) Tissue-engineered ray with (C) four
layers of body architecture, (D) concept, and (E) phototactic control. Upon optical stimulation, the tissueengineered ray induces sequential muscle activation via serpentine-patterned muscle tissues, generates
undulatory locomotion, and sustains steady forward swimming. It changes direction by generating
asymmetric undulating motion between left and right fins, modulated by light pulse frequency.
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016



Optical stimulation of the circuits initiated sequential activation waves that propagated along
the anterior-posterior axis as revealed by means
of calcium imaging (25) for three different circuit
designs (Fig. 2D, fig. S6, and movies S3 and S4).
Dense serpentine patterns enhanced activation
localization, whereas rarefied ones increased propagation speed (figs. S7 and S8). To allow for maneuverability, we ensured that each pectoral fin
could be independently actuated by pacing left
and right optical stimuli at different frequencies
(fig. S6 and movie S5). Among the various designs
considered, the circuit of choice was characterized
by the intermediate serpentine pattern density
(Fig. 2E). This circuit represented the best tradeoff
between overlap with batoids operating range
and contraction time reproducibility (low standard
error) in order to minimize desynchronization and

undesired turning (Fig. 2E). The final overall design of our ray consists of ~200,000 live cardiomyocytes in an elastomeric body of 16.3 mm
length and 10.18 0.43 mg mass.
When immersed in a 37C Tyrodes physiological salt solution containing glucose as energy
reservoir, and upon optical stimulation, the fabricated ray was propelled by producing forward
thrust via the undulatory motion of its fins (Fig. 3
and movie S6). Video-tracking analysis (25) showed
that during each swimming cycle, as the calcium
signal propagated (fig. S9), the anterior region bent
downward, while the posterior one lifted upward
(Fig. 3, A and B), conferring a small downward
orientation (10) to the rays longitudinal axis
(fig. S9). Both regions reached their maximum
displacement around ~200 ms (Fig. 3B), when
their motion gradually inverted until ~340 ms





20 mm

50 m

500 m

5 m




2 mm

500 m

50 m
5 m

without serpentine

intermediate serpentine

dense serpentine
number of traveling waves


activation delay (ms)

activation delay (ms)

activation delay (ms)

Fig. 2. Engineering solutions. (A) System-level design for skate (top) and
tissue-engineered ray (bottom left) comparable with one penny and a two
Euros coin (bottom middle and right). (B and C) Musculoskeletal (B) mesoand (C) micro-architecture of a skate, L. erinacea (top), is replicated in a tissueengineered ray (bottom). Horizontal sections of the skate were stained with
hematoxylin and eosin (top), and the engineered tissue was immunostained with
a light-sensitive membrane protein, ChR2 (red, bottom left), sarcomeric a-actinin
[red, (B), bottom right, and (C)], and nuclei (blue, (B), bottom right). (C) Orientation of the Z-lines (skate, black arrow and black triangles, top; and tissueengineered ray, black arrow with gray distribution and white triangles, bottom). In


8 JULY 2016 VOL 353 ISSUE 6295

design of

stimulation frequency (Hz)

both cases, Z-lines are perpendicular to the skeleton rays (pink arrows). (D) Muscle
circuits with preprogrammed activation pattern. A point light stimulus directed at
the front of the fins with 1.5 Hz frequency triggers the calcium wave that propagates along the predefined serpentine patterns. (E) Operating range of the
muscle circuits. The circuit with intermediate serpentine pattern density represents the best tradeoff between contraction time reproducibility (SEM) and
overlap with batoids operating range [Taeniura lymma (20) and Potamotrygon
orbignyi (22), black symbols]. Black, red, and blue indicates muscle circuit without serpentine patterns and with intermediate and dense serpentine patterns,
respectively. Each colored band indicates SEM of number of traveling waves. SCIENCE

Downloaded from on July 8, 2016

stimuli. This design was implemented by overlaying onto anisotropic tissue a layer of cardiomyocytes that were electrically coupled with
gap junctions (fig. S4H). These myocytes were
engineered to respond to optical stimuli (Fig.
1D) by expressing a light-sensitive ion channel
[channelrhodopsin-2 (ChR2)] (27). Thus, a point
light stimulus directed at the front of the ray triggers the propagation of an action potential that
is spatially and temporally modulated by the gap
junctions between muscle cells without the need
for neural coupling and coordination (Fig. 2, D
and E). The expression of ChR2 was obtained via
lentiviral transduction mediated by the truncated
cardiac troponin T promoter cTnT (fig. S5) (25, 28).
This approach led to an 88% transduction rate
of cardiomyocytes and maximized the sensitivity
to blue light at powers of ~10 mW (fig. S5).

(Fig. 3, C and D). At this point, the calcium wave

had traveled ~70% of the ray body and approached
the flexible tail region (Fig. 3E and fig. S9). This
signal caused a rapid, strong downward contraction of the rear of the body, a quick upward recoil
of the head region (presumably mediated by the
stiff skeleton) (Fig. 3), a 30 upward reorientation
of the longitudinal axis (fig. S9), and a spike in
forward swimming speed (Fig. 3E). After 340 ms,

the calcium wave vanishes, and the ray relaxes

(Fig. 3E and fig. S9), gliding until its forward
momentum dissipates.
A periodical optical stimulation leads to a
rhythmically sustained forward displacement
(Fig. 3F). To test the benefits of spatiotemporally
modulated undulatory locomotion relative to pulsatile locomotion (typical of jellyfish), we stimulated
our design of choice via a global electrical field

Fig. 3. Kinematics and hydrodynamics. (A to D) PIV flow measurements highlight the production
of alternated positive and negative vortices by the tissue-engineered ray.The viscosity associated with the
relatively small Re is responsible for the dissipation of the vortex street in the wake. (E) Correlation
between calcium activity and undulatory locomotion. (F) The moving distance during four strokes.
(G) Comparison of swimming speed between the tissue-engineered rays stimulated by point and
field stimulations. Undulatory locomotion produced by sequential muscle activation (point, 1.85 mm/s)
improved swimming speed significantly compared with pulsatile propulsion generated by global
muscle activation (field, 1.03 mm/s; matched pairs test, P = 0.014, n = 3 rays). Gray and red lines
indicate the speed of individual rays and their average, respectively. (H and I) Out-of-plane fin deflection in both (H) a live stingray, P. orbignyi, and (I) a tissue-engineered ray (maximum amplitude, 2.54
0.02 mm). (J) Comparison of swimming performance between tissue-engineered rays (n = 7 rays) and
aquatic swimmers [batoid fish (20) and larval zebrafish (32)] following the scaling analysis (23).
[Figure adapted from (23).] (K to N) PIV analysis of live Little skate, L. erinacea.

(25). The latter, unlike optical point stimulation,

induces a global, synchronized contraction of the
entire muscle layer, leading to jet-like propulsion
(movie S7). Although pulsatile actuation also produced forward motion, undulatory gaits were
found to be twice as fast at 1.5 Hz pacing (Fig. 3G).
We compared the kinematics (Fig. 3, H and I)
and hydrodynamics (Fig. 3, J to M) of bioinspired
and live rays. Both rays exhibit asymmetric deformation patterns in which the deflection amplitude progressively increases in the radially
outward and anterior-posterior directions. This
similarity shows that our asymmetric composite
structure compensates for the lack of the upper
muscle layer (Fig. 3, B and D), leading to the
coordinated undulatory locomotion. Our design
of choice was found to outperform a symmetric
design (by 5.7, as measured by distance traveled
per unit time) as well as asymmetric designs
without gold skeleton (2.7), with denser gold
skeleton (8.2), and with thinner (1.5) and thicker
fins (3.3) (fig. S10 and movie S8 and S9). Particle
image velocimetry (PIV) (25) images of the hydrodynamic footprint of the tissue-engineered
ray (Fig. 3, A to D) show that body contractions
generate vortices of alternating sign that are
sequentially shed downstream in the wake (Fig.
3, A to D, and movie S10), which is the hallmark
of inertial undulatory swimming (29). Indeed, PIV
of live skates (Fig. 3, K to M, and fig. S11) reveals a
qualitatively similar alternation of positive and
negative vortices, respectively generated in concave
and convex regions of the body.
We emphasize here that our ray is a 10-fold
scaled-down version of a live skate and moves in
a laminar flow regime, as opposed to batoid fish
that operate in turbulent conditions. Thus, a direct performance comparison is not meaningful,
but it is instructive to contextualize artificial and
natural solutions in terms of a recent scaling
framework (23, 30). All inertial undulatory swimmers hew to two scaling laws, Re = Sw4/3 in the
laminar regime and Re = Sw in the turbulent regime, where the Swimming number Sw = 2pfAL/u
(L is the characteristic length of the swimmer, f is
the undulation frequency, A is the amplitude, and
u is the fluid viscosity) captures input kinematics,
and the Reynolds number Re = UL/u (where U is
the forward speed) captures output speed. By
fitting extant biological data (23), the average
swimming roofline was determined, thus providing an objective way to assess swimming performance. In this analysis, our tissue-engineered
rays reached up to 63% of the Re of comparable
natural solutions at the given Sw (Fig. 3J).
For gait control, we first determined a set of
gait protocols for speed and direction control
by modulating light frequency and by synchronously or asynchronously triggering the right and
left serpentine circuits (figs. S11 and S12 and movies
S12 to S18). Synchronous pacing on both fins resulted in straight, forward displacement (figs. S11
and S12 and movies S12 and S13), whereas stimulation frequency determined the swimming speed
range (maximum at 1.5 to 2 Hz, minimum at 1 or
3 Hz) (fig. S10E and movies S14 to S16). Asynchronous pacing instead resulted in directional
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 4. Phototactic steering of the tissue-engineered ray through an obstacle course. (A) The ray completed a course that required complex
coordination and maneuvering. Motion began with a forward protocol (at 0 s) to gain acceleration. A following left turn protocol allowed the ray to overcome
forward momentum, making a left turn (at ~50 s). Next, another forward protocol was used to dissipate counterclockwise angular momentum and regain
directionality (at ~100 s). While the ray made its way back to the other side of the obstacles, a final right turn protocol was given to make a right turn, winding the
last obstacle. Grids, 1 cm. (B) Corresponding kinematic analysis relative to light frequency modulation protocols used for guidance.

turns (fig. S12, B, C, and F). In order to minimize

the turning radius, we paired stimulation frequencies (1/1.5 Hz in movie S17, or 3/1.5 Hz in movie S18)
to maximize the actuation difference between fins.
Our ray turned in either clockwise or counterclockwise directions by generating asymmetric
undulating motion between left and right fins, as
in batoid fishes (24).
Last, we challenged the tissue-engineered ray
to swim through an obstacle course. Using the
above gait and turning protocols, we guided the
ray along a curved path by alternating forward
motion and turning maneuvers, at an average
speed of ~1.5 mm/s over a distance of ~250 mm,
15 times longer than its body length (Fig. 4 and
movie S19). Furthermore, the ray was found able
to maintain 80% of its initial speed for up to 6
consecutive days (fig. S13 and movie S20). Therefore, our ray outperformed existing locomotive
biohybrid systems in terms of speed [3.2 mm/s in
movie S16, 1.3 over jellyfish (18)], distance traveled [~250 mm, 35 over cantilever-like walkers
(15)], and durability (6 days), demonstrating the
potential of self-propelled, phototactically activated tissue-engineered robots.
With dissociated cells, naturally equipped with
biosensors and bioactuators, as a programmable,
actuating building material, we used optogenetics
and tissue engineering to build an adaptive swimming animal. Our study is but a first step in
engineering multilevel systems that link neurodynamics, mechanics, and complex controllable
gaitscoupling sensory information to motor coordination and movement that leads to behavior.
This work paves the way for the development of
autonomous and adaptive artificial creatures able
to process multiple sensory inputs and produce
complex behaviors in distributed systems and may

8 JULY 2016 VOL 353 ISSUE 6295

represent a path toward soft-robotic embodied

cognition (4, 31).


R. Pfeifer, M. Lungarella, F. Iida, Science 318, 10881093 (2007).

M. Haberland, S. Kim, Bioinspir. Biomim. 10, 016010 (2015).
S. Kim, C. Laschi, B. Trimmer, Trends Biotechnol. 31, 287294 (2013).
D. Rus, M. T. Tolley, Nature 521, 467475 (2015).
K. Y. Ma, P. Chirarattananon, S. B. Fuller, R. J. Wood, Science
340, 603607 (2013).
J. S. Koh et al., Science 349, 517521 (2015).
M. S. Triantafyllou, G. S. Triantafyllou, Sci. Am. 272, 6470 (1995).
K. W. Moored, F. E. Fish, T. H. Kemp, H. Bart-Smith, Mar.
Technol. Soc. J. 45, 99109 (2011).
H. C. Astley et al., Proc. Natl. Acad. Sci. U.S.A. 112, 62006205 (2015).
A. J. Ijspeert, A. Crespi, D. Ryczko, J. M. Cabelguen, Science
315, 14161420 (2007).
D. J. Hyun, S. Seok, J. Lee, S. Kim, Int. J. Robot. Res. 33,
14171445 (2014).
V. Chan, H. H. Asada, R. Bashir, Lab Chip 14, 653670 (2014).
J. D. Sander, J. K. Joung, Nat. Biotechnol. 32, 347355 (2014).
A. W. Feinberg et al., Science 317, 13661370 (2007).
V. Chan et al., Sci Rep 2, 857 (2012).
R. Raman et al., Proc. Natl. Acad. Sci. U.S.A. 113, 34973502 (2016).
B. J. Williams, S. V. Anand, J. Rajagopalan, M. T. A. Saif, Nat.
Commun. 5, 3081 (2014).
J. C. Nawroth et al., Nat. Biotechnol. 30, 792797 (2012).
V. Braitenberg, Vehicles: Experiments in Synthetic Psychology
(MIT Press, 1986).
L. J. Rosenberger, M. W. Westneat, J. Exp. Biol. 202,
35233539 (1999).
V. Di Santo, C. P. Kenaley, J. Exp. Biol. 219, 18041807 (2016).
E. L. Blevins, G. V. Lauder, J. Exp. Biol. 215, 32313241 (2012).
M. Gazzola, M. Argentina, L. Mahadevan, Nat. Phys. 10,
758761 (2014).
J. M. Parson, F. E. Fish, A. J. Nicastro, J. Exp. Mar. Biol. Ecol.
402, 1218 (2011).
Materials and methods are available as supplementary
materials on Science Online.
M. Gazzola, M. Argentina, L. Mahadevan, Proc. Natl. Acad. Sci.
U.S.A. 112, 38743879 (2015).
E. S. Boyden, F. Zhang, E. Bamberg, G. Nagel, K. Deisseroth,
Nat. Neurosci. 8, 12631268 (2005).
H. Ma, C. M. Sumbilla, I. K. G. Farrance, M. G. Klein, G. Inesi,
Am. J. Physiol. Cell Physiol. 286, C556C564 (2004).

29. W. M. van Rees, M. Gazzola, P. Koumoutsakos, J. Fluid Mech.

722, R3 (2013).
30. A. J. Cote, P. W. Webb, Integr. Comp. Biol. 55, 662672 (2015).
31. R. Pfeifer, H. G. Marques, F. Iida, Soft robotics: The next
generation of intelligent machines. Proceedings of the TwentyThird International Joint Conference on Artificial Intelligence,
39 August 2013, Beijing, China, pp. 511 (2013).
32. U. K. Mller, J. L. van Leeuwen, J. Exp. Biol. 207, 853868 (2004).

Data reported in the paper are included in the supplementary materials.

We thank A. P. Nesmith, M. McKenna, and A. Chandler for discussion on
design and K. Hudson and Margherita Gazzola for photography and
illustrations. This work was funded by the Harvard Paulson School of
Engineering and Applied Sciences, the Wyss Institute for Biologically
Inspired Engineering, the National Center for Advancing Translational
Sciences grant UH3TR000522, subcontract 312659 from Los Almos
National Laboratory under prime DTRA contract DE-AC52-06NA25396,
the National Science Foundation (NSF) grant EFRI-0938043, NSF
Materials Research Science and Engineering Center grant DMR1420570, Office of Naval Research Multidisciplinary University
Research Initiative grant N000141410533, the Swiss National Science
Foundation, the MacArthur Foundation, the Radcliffe Institute, the
National Research Foundation of Korea grant 2013K1A4A3055268,
and the U.S. Army Research Laboratory and Office contract
W911NF-12-2-0036. The views and conclusions contained in this Report
are those of the authors and should not be interpreted as representing
official policies, expressed or implied, of the Army Research Office
or Laboratory, the U.S. government, or any other funding agencies.
The U.S. government is authorized to reproduce and distribute reprints
for government purposes notwithstanding any copyright notation
hereon. Portions of this work were performed at the Harvard Center for
Nanoscale Systems (NSF 1541959) and at the Neurobiology Imaging
Facility (NINDS P30 Core Center NS072030). Certain aspects of the
paper are described in U.S. patent 8,492,150 and U.S. patent
application 20150182679.
Materials and Methods
Supplementary Text
Figs. S1 to S15
References (3344)
Movies S1 to S20
6 February 2016; accepted 19 May 2016
10.1126/science.aaf4292 SCIENCE

Downloaded from on July 8, 2016



Higher-order organization of
complex networks
Austin R. Benson,1 David F. Gleich,2 Jure Leskovec3*
Networks are a fundamental tool for understanding and modeling complex systems in physics,
biology, neuroscience, engineering, and social science. Many networks are known to exhibit rich,
lower-order connectivity patterns that can be captured at the level of individual nodes and
edges. However, higher-order organization of complex networksat the level of small network
subgraphsremains largely unknown. Here, we develop a generalized framework for clustering
networks on the basis of higher-order connectivity patterns. This framework provides
mathematical guarantees on the optimality of obtained clusters and scales to networks with
billions of edges. The framework reveals higher-order organization in a number of networks,
including information propagation units in neuronal networks and hub structure in transportation
networks. Results show that networks exhibit rich higher-order organizational structures
that are exposed by clustering based on higher-order connectivity patterns.

etworks are a standard representation of

data throughout the sciences, and higherorder connectivity patterns are essential to
understanding the fundamental structures
that control and mediate the behavior of
many complex systems (17). The most common
higher-order structures are small network subgraphs, which we refer to as network motifs (Fig. 1A).
Network motifs are considered building blocks
for complex networks (1, 8). For example, feedforward loops (Fig. 1A, M5) have proven fundamental to understanding transcriptional regulation
networks (9); triangular motifs (Fig. 1A, M1M7) are
crucial for social networks (4); open bidirectional
wedges (Fig. 1A, M13) are key to structural hubs
in the brain (10); and two-hop paths (Fig. 1A,
M8M13) are essential to understanding air traffic patterns (5). Although network motifs have
been recognized as fundamental units of networks, the higher-order organization of networks
at the level of network motifs largely remains an
open question.
Here, we use higher-order network structures
to gain new insights into the organization of complex systems. We develop a framework that identifies clusters of network motifs. For each network
motif (Fig. 1A), a different higher-order clustering
may be revealed (Fig. 1B), which means that different organizational patterns are exposed, depending on the chosen motif.
Conceptually, given a network motif M, our
framework searches for a cluster of nodes S with
two goals. First, the nodes in S should participate
in many instances of M. Second, the set S should
avoid cutting instances of M, which occurs when
only a subset of the nodes from a motif are in the
set S (Fig. 1B). More precisely, given a motif M,
the higher-order clustering framework aims to
find a cluster (defined by a set of nodes S) that

Institute for Computational and Mathematical Engineering,

Stanford University, Stanford, CA 94305, USA. 2Department
of Computer Science, Purdue University, West Lafayette, IN
47906, USA. 3Computer Science Department, Stanford
University, Stanford, CA 94305, USA.
*Corresponding author. Email:


minimizes the following ratio:

fM S cutM S; S=minvolM S; volM S
where S denotes the remainder of the nodes (the
complement of S), cutM(S,S) is the number of
instances of motif M with at least one node in S
and one in S, and volM (S) is the number of nodes

in instances of M that reside in S. Equation 1 is a

generalization of the conductance metric in spectral graph theory, one of the most useful graph
partitioning scores (11). We refer to fM(S) as the
motif conductance of S with respect to M.
Finding the exact set of nodes S that minimizes
the motif conductance is computationally infeasible
(12). To approximately minimize Eq. 1 and, hence,
to identify higher-order clusters, we developed an
optimization framework that provably finds nearoptimal clusters [supplementary materials (13)].
We extend the spectral graph clustering methodology, which is based on the eigenvalues and eigenvectors of matrices associated with the graph (11),
to account for higher-order structures in networks.
The resulting method maintains the properties of
traditional spectral graph clustering: computational
efficiency, ease of implementation, and mathematical guarantees on the near-optimality of obtained
clusters. Specifically, the clusters identified by our
higher-order clustering framework satisfy the motif
Cheeger inequality (14), which means that our
optimization framework finds clusters that are at
most a quadratic factor away from optimal.
The algorithm (illustrated in Fig. 1C) efficiently
identifies a cluster of nodes S as follows:
Step 1: Given a network and a motif M of
interest, form the motif adjacency matrix WM
whose entries (i, j) are the co-occurrence counts
of nodes i and j in the motif M: (WM)ij = number
of instances of M that contain nodes i and j.

Fig. 1. Higher-order network structures and the higher-order network clustering framework.
(A) Higher-order structures are captured by network motifs. For example, all 13 connected three-node directed
motifs are shown here. (B) Clustering of a network based on motif M 7. For a given motif M, our framework aims to
find a set of nodes S that minimizes motif conductance, fM(S), which we define as the ratio of the number of
motifs cut (filled triangles cut) to the minimum number of nodes in instances of the motif in either S or S (13). In
this case, there is one motif cut. (C) The higher-order network clustering framework. Given a graph and a motif of
interest (in this case, M7), the framework forms a motif adjacency matrix (WM) by counting the number of times
two nodes co-occur in an instance of the motif. An eigenvector of a Laplacian transformation of the motif
adjacency matrix is then computed.The ordering s of the nodes provided by the components of the eigenvector
(15) produces nested sets Sr = {s1, , sr} of increasing size r. We prove that the set Sr with the smallest motifbased conductance, fM(Sr), is a near-optimal higher-order cluster (13).
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 2. Higher-order cluster in the C. elegans neuronal network. [See (29).]

(A) The four-node bi-fan motif, which is overexpressed in neuronal networks
(1). Intuitively, this motif describes a cooperative propagation of information
from the nodes on the left to the nodes on the right. (B) The best higher-order
cluster in the C. elegans frontal neuronal network based on the motif in (A).
The cluster contains three ring motor neurons (RMEL, -V, and -R; cyan) with
many outgoing connections, which serve as the source of information; six
inner labial sensory neurons (IL2DL, -VR, -R, -DR, -VL, and -L; orange) with
many incoming connections, serving as the destination of information; and four
URA motor neurons (purple) acting as intermediaries. These RME neurons

Step 2: Compute the spectral ordering s of

the nodes from the normalized motif Laplacian
matrix constructed via WM (15).
Step 3: Find the prefix set of s with the
smallest motif conductance (the argument of the
minimum), formally, S : = arg minr fM(Sr), where
Sr = {s1, , sr}.
For triangular motifs, the algorithm scales to
networks with billions of edges and, typically, only
takes several hours to process graphs of such size.
On smaller networks with hundreds of thousands
of edges, the algorithm can process motifs up to
size 9 (13). Although the worst-case computational
complexity of the algorithm for triangular motifs is Q(m1.5), where m is the number of edges

8 JULY 2016 VOL 353 ISSUE 6295

have been proposed as pioneers for the nerve ring (21), whereas the IL2
neurons are known regulators of nictation (22), and the higher-order cluster
exposes their organization.The cluster also reveals that RIH serves as a critical
intermediary of information processing. This neuron has incoming links from
three RME neurons, outgoing connections to five of the six IL2 neurons, and
the largest total number of connections of any neuron in the cluster. (C) Illustration of the higher-order cluster in the context of the entire network. Node
locations are the true two-dimensional spatial embedding of the neurons.
Most information flows from left to right, and we see that RMEV, -R, and -L and
RIH serve as sources of information to the neurons on the right.

in the network, in practice, the algorithm is much

faster. By analyzing 16 real-world networks where
the number of edges m ranges from 159,000 to
2 billion, we found the computational complexity to scale as Q(m1.2). Moreover, the algorithm
can easily be parallelized, and sampling techniques can be used to further improve performance (16).
The framework can be applied to directed, undirected, and weighted networks, as well as motifs
(13). Moreover, it can also be applied to networks
with positive and negative signs on the edges, which
are common in social networks (friend versus foe
or trust versus distrust edges) and metabolic networks (edges signifying activation versus inhibi-

tion) (13). The framework can be used to identify

higher-order structure in networks where domain
knowledge suggests the motif of interest. In
the supplementary materials, we also show that
when a domain-specific higher-order pattern
is not known in advance, the framework can
also serve to identify which motifs are important
for the modular organization of a given network
(13). Such a general framework allows complex
higher-order organizational structures in a number of different networks by using individual
motifs and sets of motifs. The framework and
mathematical theory immediately extend to other
spectral methods, such as localized algorithms
that find clusters around a seed node (17) and SCIENCE

Downloaded from on July 8, 2016


Fig. 3. Higher-order spectral analysis of a network

of airports in Canada and
the United States. [See
(23).] (A) The three higherorder structures used in our
analysis. Each motif is
anchored by the blue nodes
i and j, which means our
framework only seeks to
cluster together the blue
nodes. Specifically, the motif
adjacency matrix adds weight
to the (i, j) edge on the basis
of the number of third intermediary nodes (green
squares).The first two motifs
correspond to highly connected cities, and the motif
on the right connects nonhubs to nonhubs. (B) The top
50 most populous cities in
the United States, which
correspond to nodes in the
network.The edge thickness
is proportional to the weight
in the motif adjacency matrix
WM. The thick, dark lines
indicate that large weights
correspond to popular
mainline routes.
(C) Embedding of nodes
provided by their
corresponding components
of the first two nontrivial eigenvectors of the normalized Laplacian for WM.The marked cities are eight large U.S. hubs (green), three West Coast nonhubs (red), and
three East Coast nonhubs (purple).The primary spectral coordinate (left to right) reveals how much of a hub the city is, and the second spectral coordinate (top to
bottom) captures west-east geography (13). (D) Embedding of nodes provided by their corresponding components in the first two nontrivial eigenvectors of
the standard, edge-based (nonhigher-order) normalized Laplacian. This method does not capture the hub and geography found by the higher-order
method. For example, Atlanta, the largest hub, is in the center of the embedding, next to Salina, a nonhub.

algorithms for finding overlapping clusters (18).

To find several clusters, one can use embeddings
from multiple eigenvectors and k-means clustering (13, 19) or can apply recursive bipartitioning (13, 20).
The framework can serve to identify a higher-order
modular organization of networks. We apply the
higher-order clustering framework to the Caenorhabditis elegans neuronal network, where the
four-node bi-fan motif (Fig. 2A) is overexpressed
(1). The higher-order clustering framework then
reveals the organization of the motif within the C.
elegans neuronal network. We find a cluster of
20 neurons in the frontal section with low bi-fan
motif conductance (Fig. 2B). The cluster shows
a way that nictation is controlled. Within the cluster, ring motor neurons (RMEL, -V, or -R), proposed pioneers of the nerve ring (21), propagate
information to inner labial sensory neurons,
regulators of nictation (22), through the neuron
RIH (Fig. 2C). Our framework contextualizes the
importance of the bi-fan motif in this control
The framework also provides new insights into
network organization beyond the clustering of
nodes based only on edges. Results on a transSCIENCE

portation reachability network (23) demonstrate

how it finds the essential hub interconnection
airports (Fig. 3). These appear as extrema on
the primary spectral direction (Fig. 3C) when twohop motifs (Fig. 3A) are used to capture highly
connected nodes and nonhubs. [The first spectral
coordinate of the normalized motif Laplacian embedding was positively correlated with the airport
citys metropolitan population with Pearson correlation 99% confidence interval (0.33, 0.53).] The
secondary spectral direction identified the west-east
geography in the North American flight network
[it was negatively correlated with the airport citys
longitude with Pearson correlation 99% confidence
interval (0.66, 0.50)]. On the other hand, edgebased methods conflate geography and hub structure. For example, Atlanta, a large hub, is embedded
next to Salina, a nonhub, with an edge-based
method (Fig. 3D).
Our higher-order network clustering framework
unifies motif analysis and network partitioning
two fundamental tools in network scienceand
reveals new organizational patterns and modules
in complex systems. Prior efforts along these lines
do not provide worst-case performance guarantees
on the obtained clustering (24) and do not reveal

which motifs organize the network (25) but rely

on expanding the size of the network (26, 27).
Theoretical results in the supplementary materials (13) also explain why classes of hypergraph
partitioning methods are more general than previously assumed and how motif-based clustering
provides a rigorous framework for the special
case of partitioning directed graphs. Finally, the
higher-order network clustering framework is
generally applicable to a wide range of network
types, including directed, undirected, weighted,
and signed networks.

1. R. Milo et al., Science 298, 824827 (2002).

2. S. Mangan, A. Zaslaver, U. Alon, J. Mol. Biol. 334, 197204
3. J. Yang, J. Leskovec, Proc. IEEE 102, 18921902 (2014).
4. P. W. Holland, S. Leinhardt, Am. J. Sociol. 76, 492513
5. M. Rosvall, A. V. Esquivel, A. Lancichinetti, J. D. West,
R. Lambiotte, Nat. Commun. 5, 4630 (2014).
6. N. Prulj, D. G. Corneil, I. Jurisica, Bioinformatics 20,
35083515 (2004).
7. J. Leskovec, K. J. Lang, A. Dasgupta, M. W. Mahoney, Internet
Math. 6, 29123 (2009).
8. . N. Yaverolu et al., Sci. Rep. 4, 4547 (2014).
9. S. Mangan, U. Alon, Proc. Natl. Acad. Sci. U.S.A. 100,
1198011985 (2003).

8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016




inverse square roots on the diagonal Dii







1= Sj W M ij . The

spectral ordering s is the by-value ordering of D1/2z, where z is

the eigenvector corresponding to the second smallest
eigenvalue of LM, i.e., si is the index of D1/2z with the ith
smallest value.
C. Seshadhri, A. Pinar, T. G. Kolda, Stat. Anal. Data Min. 7,
294307 (2014).
R. Andersen, F. Chung, K. Lang, in Proceedings of the 47th
Annual IEEE Symposium on Foundations of Computer Science,
FOCS06, Berkeley, California, 21 to 25 October 2006 (Institute
of Electrical and Electronics Engineers, Piscataway, NJ, 2006),
pp. 475486.
J. J. Whang, I. S. Dhillon, D. F. Gleich, in Proceedings of the
2015 SIAM International Conference on Data Mining, Vancouver,
British Columbia, Canada, 30 April to 2 May 2015,
S. Venkatasubramanian, J. Ye, Eds. (Society for Industrial and
Applied Mathematics, Philadelphia, PA, 2015), pp. 936944.
A. Y. Ng, M. I. Jordan, Y. Weiss, Adv. Neural Inf. Process. Syst.
14, 849856 (2002).
D. Boley, Data Min. Knowl. Discov. 2, 325344 (1998).
D. L. Riddle et al., Eds., C. elegans II (Cold Spring
Harbor Laboratory Press, Cold Spring Harbor, NY,
ed. 2, 1997).
H. Lee et al., Nat. Neurosci. 15, 107112 (2011).
B. J. Frey, D. Dueck, Science 315, 972976
B. Serrour, A. Arenas, S. Gmez, Comput. Commun. 34,
629634 (2011).
T. Michoel, A. Joshi, B. Nachtergaele, Y. Van de Peer, Mol.
Biosyst. 7, 27692778 (2011).
A. R. Benson, D. F. Gleich, J. Leskovec, in Proceedings of the
2015 SIAM International Conference on Data Mining, Vancouver,
British Columbia, Canada, 30 April to 2 May 2015,
S. Venkatasubramanian, J. Ye, Eds. (SIAM, Philadelphia, 2015),
pp. 118126.
F. Krzakala et al., Proc. Natl. Acad. Sci. U.S.A. 110,
2093520940 (2013).
D. Wagner, F. Wagner, in Mathematical Foundations
of Computer Science 1993, Proceedings of the
18th International Symposium on Mathematical Foundations
of Computer Science, MFCS93, Gdask, Poland, 30 August to
3 September 1993, A. M. Borzyszkowski, S. Sokolowski, Eds.
(Lecture Notes in Computer Science, Springer, New York,
1993), pp. 744750.
M. Kaiser, C. C. Hilgetag, PLOS Comput. Biol. 2, e95


The authors thank R. Sosi for insightful comments. A.R.B. was

supported by a Stanford Graduate Fellowship; D.F.G. was
supported by NSF (CCF-1149756 and IIS-1422918), J.L. was
supported by NSF (IIS-1149837 and CNS-1010921), trans-NIH
initiative Big Data to Knowledge (BD2K), Defense Advanced
Research Projects Agency [XDATA and Simplifying Complexity in
Scientific Discovery (SIMPLEX)], Boeing, Lightspeed, and
Volkswagen. Software implementations and the data sets used
to obtain the results in this manuscript are available at
Materials and Methods
Supplementary Text
Figs. S1 to S13
Tables S1 to S12
References (3084)
18 November 2015; accepted 18 May 2016


8 JULY 2016 VOL 353 ISSUE 6295


S-Acylation of the cellulose synthase

complex is essential for its plasma
membrane localization
Manoj Kumar,1 Raymond Wightman,2 Ivan Atanassov,1* Anjali Gupta,1
Charlotte H. Hurst,3,4 Piers A. Hemsley,3,4 Simon Turner1
Plant cellulose microfibrils are synthesized by a process that propels the cellulose
synthase complex (CSC) through the plane of the plasma membrane. How interactions
between membranes and the CSC are regulated is currently unknown. Here, we
demonstrate that all catalytic subunits of the CSC, known as cellulose synthase A (CESA)
proteins, are S-acylated. Analysis of Arabidopsis CESA7 reveals four cysteines in variable
region 2 (VR2) and two cysteines at the carboxy terminus (CT) as S-acylation sites.
Mutating both the VR2 and CT cysteines permits CSC assembly and trafficking to the Golgi
but prevents localization to the plasma membrane. Estimates suggest that a single CSC
contains more than 100 S-acyl groups, which greatly increase the hydrophobic nature of
the CSC and likely influence its immediate membrane environment.

ellulose in plants is synthesized at the plasma membrane by the cellulose synthase

complex (CSC), which contains at least 18
catalytic CESA protein subunits (1). The
direction of CSC movement and the orientation of cellulose microfibril deposition are determined by cortical microtubules (2). Movement of
the CSC through the plane of the plasma membrane is likely to cause severe disruption to the
lipid bilayer (3), which suggests that membrane
partitioning of this process may be important.
Here, we describe the modifications of CESA proteins and demonstrate their importance to the
functioning of the CSC.
S-Acylation involves reversible addition of an
acyl group, often palmitate or stearate, to a cysteine residue, which can affect protein structure
or localization (4). A recent study identified many
S-acylated proteins in plants (5), including CESA1
and CESA3, which are essential for cellulose
synthesis in the primary cell wall (6). We used
acylresin-assisted capture (acyl-RAC) assays (7)
to confirm that CESA1 is S-acylated (fig. S1) and
showed that CESA6 is also S-acylated (Fig. 1A).
Furthermore, all three CESAs required for cellulose synthesis in the secondary cell wall, CESA4,
CESA7, and CESA8, are S-acylated (Fig. 1A), which
demonstrates that S-acylation is a common feature of CESA proteins involved in cellulose synthesis in both primary and secondary cell walls.
CESA7 has 26 cysteines (fig. S2A). In order to
identify S-acylated cysteines, we mutated indi1

Faculty of Life Sciences, The University of Manchester, Michael

Smith Building, Oxford Road, Manchester M13 9PT, UK.
Microscopy Core Facility, Sainsbury Laboratory, University of
Cambridge, Bateman Street, Cambridge CB2 1LR, UK. 3Division
of Plant Sciences, School of Life Sciences, University of
Dundee, Dow Street, Dundee, DD1 5EH, Scotland, UK. 4Cell and
Molecular Sciences, The James Hutton Institute, Invergowrie,
DD2 5DA, Scotland, UK.

*Present address: AgroBioInstitute, 8 Dragan Tzankov Boulevard,

1164 Sofia; Bulgaria. Corresponding author. Email: simon.turner@ (S.T.); (P.A.H.)

vidual CESA7 cysteines to serines and tested their

ability to complement the cesa7 irx3-1 mutant. None
of the eight cysteines in the zinc finger domain
(ZR) showed any significant complementation
(Fig. 2A and figs. S3 and S4). The structure of the
RING-type zinc-finger domain from CESA7 [Protein Data Bank (PDB) ID: 1WEO] shows that all
eight cysteines are involved in coordinating two
zinc atoms, which makes them unlikely to be
S-acylated. Consequently, we focused our subsequent analysis on other regions of CESA7. Two
highly conserved cysteines in the short C terminus (table S1) are also essential for CESA protein
function (Fig. 2A). None of the remaining 16 single cysteine mutants showed a substantial effect
on cellulose content (Fig. 2A).
A cysteine-rich region lies within VR2 (8). The
number of VR2 cysteines is conserved among
orthologous CESAs from different species but
varies between paralogous CESAs (table S1). There
are four VR2 cysteines in CESA7 (fig. S2), and
mutating them individually has no effect on cellulose biosynthesis (Fig. 2, A and C). We hypothesized that if VR2 is a site of CESA S-acylation, the
remaining VR2 cysteines may support sufficient
S-acylation for CESA7 function. Consequently, we
mutated all four VR2 cysteines in CESA7 (VR2C/S).
The VR2C/S mutant exhibited no complementation of cesa7 irx3-1 (Fig. 2C). Thus, the cysteines in
this region appear to be functionally redundant.
Having identified the VR2 and CT cysteines as
potential S-acylation sites, we proceeded to determine if these sites were S-acylated. We generated
a mutant in which both CT cysteines were mutated
(CTC/S). The CTC/S mutant did not complement the
cesa7 irx3-1 mutant (Fig. 2B). Using Acyl-RAC assays
we consistently found that S-acylation was dramatically reduced in the VR2C/S mutant, although
some signal remained. The CTC/S mutants exhibited a smaller decrease in S-acylation (Fig. 1, B
and C). We then constructed a mutant in which
both the VR2 and CT cysteines were mutated SCIENCE

Downloaded from on July 8, 2016

10. C. J. Honey, R. Ktter, M. Breakspear, O. Sporns, Proc. Natl.

Acad. Sci. U.S.A. 104, 1024010245 (2007).
11. S. E. Schaeffer, Comput. Sci. Rev. 1, 2764 (2007).
12. Minimizing fM(S) is nondeterministic polynomial-time hard
(NP-hard), which follows from the NP-hardness of the
traditional definition of conductance (28).
13. See the supplementary materials on Science Online.
14. Formally, when the motif has three nodes, the selected cluster
S satisfies fM S 4 fM 1, where fM is the smallest motif
conductance of any possible node set S. This inequality is
proved in the supplementary materials.
15. The normalized motif Laplacian matrix is LM = D1/2(D WM)D1/2,
where D is a diagonal matrix with the row-sums of WM on the
with the
diagonal [Dii = Sj (WM)ij], and D1/2 is the same matrix

(VR2/CTC/S). The VR2/CTC/S mutant exhibited no

complementation (Fig. 2B) and CESA7 S-acylation
was effectively abolished (Fig. 1, B and C), consistent with the hypothesis that both the VR2
and CT cysteines are S-acylated.
In order to determine whether some of the VR2
cysteines were more important than the others,
these cysteines were mutated in all possible double and triple combinations. All 10 of these double and triple VR2 cysteine mutants had a lower
cellulose content than the single VR2 cysteine
mutants (Fig. 2C), which indicated that all four
cysteines are important. In order to explore sequence specificity around the VR2 cysteines, we
replaced the sixamino acidlong motif containing four cysteines of CESA7 with a motif from the
same region of either CESA8 (CESA7CESA8VRC),
which contains only three cysteines, or CESA4

(CESA7CESA4VRC), which contains six cysteines

in a 12amino acidlong motif (fig. S2B). The
CESA7CESA4VRC construct complemented the
cesa7 irx3-7 mutant, but CESA7CESA8VRC did not
(Fig. 2D and fig. S3D). We also observed that
wild-type levels of CESA7 S-acylation were restored in CESA7CESA4VRC but only partially in
CESA7CESA8VRC (fig. S5). Restoration of activity
in CESA7CESA4VRC would suggest that the lack of
acylation in the VR2C/S mutant is not a result of
changes to protein conformation caused by mutating the cysteines in this region.
To investigate the effect of S-acylation on the
trafficking of the CSC, we performed live-cell
imaging of CESA7 in the developing xylem of
intact roots (Fig. 3). During vessel differentiation,
CSCs can be seen both moving around the cell in
the Golgi, where they exhibit a characteristic ring-

Fig. 1. Analysis of Arabidopsis CESA protein S-acylation. S-Acylation of various CESA proteins from
wild-type (WT) plants assayed using biotin exchange or acyl-RAC assays (A) and CESA7 cysteine cluster
mutants measured by acyl-RAC (B). For each assay, the experimental sample (EX) was compared with
the loading control (LC) with or without (+/-) hydroxylamine (HYD) for hydroxylamine-dependent
capture of S-acylated proteins. Individual CESA proteins were detected with specific antibodies. (C) The
genotypes shown in (B) were tested in four independent experiments. Relative levels of S-acylation are
expressed as a percentage of the wild type. Error bars represent SEMs.

Fig. 2. Cellulose content of the CESA7 cysteine mutants. Cellulose content

of single cysteine mutants (A), cysteine cluster mutants (B), VR2 cysteines
mutant combinations (C), and mutants with CESA7 VR2 cysteines replaced
with those of CESA4 and CESA8 (D). The numbers on the horizontal axis in
(A) and (C) refer to the CESA7 residue number of the mutated cysteines. ZR,
VR2, and CT domains mentioned in text are highlighted with black lines in (A).


shaped morphology, and as transverse bands at

the plasma membrane that correspond to sites of
secondary cell-wall deposition and cortical microtubule localization (9, 10). Using wild-type yellow
fluorescent protein (YFP)labeled CESA7 (YFPCESA7), this banded pattern can be seen in single
images, movie projections (Fig. 3, A and C), and
movies (movie S1) and colocalizes with bands of
cortical microtubules (fig. S6). In contrast, neither
of the VR2C/S or CTC/S mutants exhibited the
banded pattern (Fig. 3, A and C, and movies S2
and S3). Furthermore, we did not observe colocalization of CESA7 with cortical microtubules in
the VR2C/S mutant (fig. S6). The banded pattern
of CESA7 localization is only apparent during
later stages of vessel differentiation. At earlier
stages, YFP-CESA7 is localized mainly within the
Golgi, a distribution that resembles those seen
in the VR2C/S and CTC/S mutants. Similarly,
mechanical perturbation of the seedlings during
the imaging process can lead to a loss of the
banded pattern. In plants that contained both
cyan fluorescent protein (CFP)tagged wild-type
CESA7 and YFP-tagged VR2C/S, we were able to
see bands with CFP-CESA7, indicative of localization at the plasma membrane, but not with
YFP-VR2C/S (Fig. 3B). Although we always observed banding in the lines containing only the
CFP-CESA7, banding was not observed in additional lines with high levels of YFP-VR2C/S expression, which indicated that this construct has a
dominant-negative effect on CFP-CESA7 localization. Trafficking of the CSC to the plasma membrane is known to occur from the Golgi via
particles known as microtubule-associated cellulose synthase compartments (MASCs) or small
CESA compartments (SmaCCs) (11, 12). We analyzed the movement of the Golgi and found no
differences in trafficking dynamics between the
wild-type and either the CESA7 VR2C/S or CTC/S
mutants (fig. S7A). The imaging data are consistent with the hypothesis that the defective

Cellulose content is expressed as percentage of the wild type by using

Landsberg erecta (Ler-0) for cesa7 irx3-1 [(A), (B), (C)] or Colombia (Col-0) for
cesa7 irx3-7 (D). Error bars are SEM. Significance values are for comparison of
each genotype to the mutant control in a univariate analysis of variance
(ANOVA) test. Significance at ***P < 0.001, **P < 0.01, or *P < 0.05 levels
are indicated.
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 3. Loss of acylation affects CSC localization at the plasma membrane. In live-cell imaging of
the CSC in developing xylem vessels of intact roots using YFP-CESA7, CSCs at the plasma membrane
appear as lateral bands (asterisks); whereas in the Golgi, CSCs are visualized as ring-shaped particles
(arrows). (A) Single-frame images showing localization of YFP-tagged wild-type CESA7 and the VR2C/S
and CTC/S mutants. (B) Single-frame images showing CFP-CESA7 (magenta) and YFP-VR2C/S (green)
within the same cell. (C) Average projections taken from movies showing YFP-tagged WT CESA7 and the
VR2C/S and CTC/S mutants. Scale bars, 5 m.

S-acylation of the VR2C/S mutant interferes with

Golgi to plasma membrane trafficking of the CSC.
In both the CESA7 VR2C/S and CTC/S mutants, we
were able to observe particles that resemble MASCs
(fig. S7B), but using the intact root system, we
were unable to track their movement or observe
individual insertion events. Consequently, we cannot exclude the possibility of S-acylationdefective
CSCs being transiently inserted into the plasma
membrane and rapidly recycled or plasma membrane insertion occurring at a much-reduced rate,
nor can we exclude the possibility that S-acylation
affects the ability of CSC complexes to associate
with cortical microtubules. In the future, it will
be interesting to determine exactly at what point
the trafficking is defective in the acylationdeficient mutant and whether interactions between the MASC and microtubules or other markers
for the sites of cell-wall deposition are altered.
To explore how the trafficking of CESA is altered in the S-acylationdeficient mutant, we examined the interactions of CESA subunits, because
mutations in one CESA protein can affect the association among the remaining CESAs in the complex (13). We used the tag on the VR2/CTC/S mutant
in a pull-down assay of CESA7 and found that both

8 JULY 2016 VOL 353 ISSUE 6295

CESA4 and CESA8 could be coprecipitated. This

suggests that all three CESAs are still able to associate as part of a complex, even in the absence
of normal CESA7 S-acylation (Fig. 4A). S-Acylation
of CESA4 and CESA8 remained intact in the CESA7
VR2/CTC/S mutant (Fig. 4B) which indicated that
an S-acylation defect in one subunit does not affect the S-acylation states of the others. Acylation
of membrane proteins often causes conformation
changes that may be essential for targeting and/or
interactions with other proteins (4); so, whereas
CESA proteins still associate in CESA7 VR2/CTC/S
mutants, potentially they may have altered confirmations that could affect the correct assembly
of the CSC.
VR2 encompasses a region known as the classspecific region (CSR) (8). Modeling studies have
suggested that the CSR loops away from the catalytic core, which makes it a candidate for subunit interactions (14). S-Acylation of VR2 cysteines
would alter the conformational prediction for this
region by placing the beginning of the CSR adjacent to the plasma membrane, which suggests
further structural data for this region is required.
The acyl groups are likely to be inserted into the
plasma membrane close to the transmembrane

Fig. 4. Loss of CESA7 acylation does not affect

the CSC subunit association or S-acylation of
other CESAs in the complex. (A) Total protein
and Ni2+ affinity purified CESA7 samples from
CESA7 VR2/CTC/S mutant were probed with the
CESA antibodies as indicated. PIP2 was used as a
control. (B) S-Acylation status of CESA4 (top) and
CESA8 (bottom) was determined in CESA7_WT
and CESA7 VR2/CTC/S plants by using Acyl-RAC.
For each assay, the experimental samples (EX) were
compared with the loading control (LC) with or without (+/-) hydroxylamine (HYD) for hydroxylaminedependent capture of S-acylated proteins. Blots
were probed with either CESA4- or CESA8-specific

region, in a region that is occupied by the bacterial cellulose synthase B (BcsB) in the BcsA/B crystal structure from Rhodobacter (6, 14, 15). One of
the major differences between the way in which
bacteria and plants make cellulose is the mobile
nature of the plant CSC, which must be able to
move through the plane of the plasma membrane as it makes cellulose. We speculate that
this process may be facilitated by the S-acylation of
plant CESA proteins.
Up to six cysteines are likely to be S-acylated in
each of CESA7, CESA4, and CESA8. Thus, a functional CSC, with at least 18 CESA proteins, would
contain more than 100 S-acyl groups. The hydrophobic nature of S-acylation can make some proteins resistant to solubilization even by ionic
detergents such as SDS (16). S-Acyl groups can be
removed by treatment with dithiothreitol (DTT)
(17), which also reduces aggregation of CESA proteins during SDS-polyacrylamide gel electrophoresis (18). Using mass spectrometry, we have only
been able to identify VR2 peptides from DTT-treated
samples. The very high level of S-acylation and hydrophobicity of the CSC and/or the requirement
for a specialized membrane environment is likely
to make the CSC susceptible to aggregation and SCIENCE

Downloaded from on July 8, 2016



may explain why it has not been possible to purify

an intact, active CSC. Although S-acylation is known
to affect partitioning of proteins into membrane
microdomains, it has also been suggested that the
crowding of S-acyl groups within a membrane
may actually facilitate the formation of lipid microdomains (19). The high level of S-acylation found in
the CSC would make it a very good candidate for a
protein complex capable of generating lipid microdomains that may facilitate the colocalization of
proteins with similar properties. We note that a
recent proteomic study of S-acylated proteins also
identified the endoglucanase KORRIGAN, a known
CESA-binding protein (5).


Climate-driven regime shift of a

temperate marine ecosystem
Thomas Wernberg,1* Scott Bennett,1,2,3 Russell C. Babcock,1,4 Thibaut de Bettignies,1,5
Katherine Cure,1,6 Martial Depczynski,6 Francois Dufois,7 Jane Fromont,8
Christopher J. Fulton,9 Renae K. Hovey,1 Euan S. Harvey,2 Thomas H. Holmes,1,10
Gary A. Kendrick,1 Ben Radford,6,11 Julia Santana-Garcon,1,2,3 Benjamin J. Saunders,2
Dan A. Smale,1,11 Mads S. Thomsen,1,12 Chenae A. Tuckett,1 Fernando Tuya,13
Mathew A. Vanderklift,7 Shaun Wilson1,10

1. A. N. Fernandes et al., Proc. Natl. Acad. Sci. U.S.A. 108,

E1195E1203 (2011).
2. A. R. Paredez, C. R. Somerville, D. W. Ehrhardt, Science 312,
14911495 (2006).
3. F. Diotallevi, B. Mulder, Biophys. J. 92, 26662673
4. P. A. Hemsley, New Phytol. 205, 476489 (2015).
5. P. A. Hemsley, T. Weimar, K. S. Lilley, P. Dupree, C. S. Grierson,
New Phytol. 197, 805814 (2013).
6. M. Kumar, S. Turner, Phytochemistry 112, 9199
7. M. T. Forrester et al., J. Lipid Res. 52, 393398
8. C. E. Vergara, N. C. Carpita, Plant Mol. Biol. 47, 145160
9. R. Wightman, S. R. Turner, Plant J. 54, 794805
10. Y. Watanabe et al., Science 350, 198203
11. E. F. Crowell et al., Plant Cell 21, 11411154
12. R. Gutierrez, J. J. Lindeboom, A. R. Paredez, A. M. C. Emons,
D. W. Ehrhardt, Nat. Cell Biol. 11, 797806 (2009).
13. N. G. Taylor, R. M. Howells, A. K. Huttly, K. Vickers, S. R. Turner,
Proc. Natl. Acad. Sci. U.S.A. 100, 14501455 (2003).
14. L. Sethaphong et al., Proc. Natl. Acad. Sci. U.S.A. 110,
75127517 (2013).
15. J. L. W. Morgan, J. Strumillo, J. Zimmer, Nature 493, 181186
16. S. Monier, D. J. Dietzen, W. R. Hastings, D. M. Lublin,
T. V. Kurzchalia, FEBS Lett. 388, 143149 (1996).
17. O. Batistic, N. Sorek, S. Schltke, S. Yalovsky, J. Kudla, Plant
Cell 20, 13461362 (2008).
18. I. I. Atanassov II, J. K. Pittman, S. R. Turner, J. Biol. Chem. 284,
38333841 (2009).
19. S. S. A. Konrad, T. Ott, Trends Plant Sci. 20, 351361

We thank J. Ogas for a critical reading of the manuscript.

H. Hfte kindly provided antibodies against CESA1 and
CESA6. The work was funded by Biotechnology and Biological
Sciences Research Council grants BB/H012923/1 and
BB/M004031/1 to S.T. and BB/M024911/1 to P.A.H. The
Microscopy Facility at the Sainsbury Laboratory is supported
by the Gatsby Charitable Foundation. The authors declare
no conflict of interest. M.K., R.W., I.A., A.G., C.H.H., and
P.A.H. carried out the experimental work. M.K., P.A.H.,
and S.T. wrote the manuscript and conceived the experiments.
Supplementary materials contain additional data. S.T.
would like to dedicate this manuscript to physicist
Roy Turner (19312016) for his support, encouragement, and help
with advanced mathematics.
Materials and Methods
Figs. S1 to S8
Tables S1 and S2
Movies S1 to S3
References (2025)
5 February 2016; accepted 6 June 2016


Ecosystem reconfigurations arising from climate-driven changes in species distributions are

expected to have profound ecological, social, and economic implications. Here we reveal a rapid
climate-driven regime shift of Australian temperate reef communities, which lost their defining
kelp forests and became dominated by persistent seaweed turfs. After decades of ocean
warming, extreme marine heat waves forced a 100-kilometer range contraction of extensive
kelp forests and saw temperate species replaced by seaweeds, invertebrates, corals, and fishes
characteristic of subtropical and tropical waters. This community-wide tropicalization
fundamentally altered key ecological processes, suppressing the recovery of kelp forests.

road-scale losses of species that provide

the foundations for habitats cause dramatic
shifts in ecosystem structure, because they
support core ecological processes (13). Such
habitat loss can lead to a regime shift, in
which reinforcing feedback mechanisms intensify
to provide resilience to an alternate community
configuration, often with profound ecological,
social, and economic consequences (46). Benthic
marine regime shifts have been associated with
the erosion of ecological resilience through overfishing or eutrophication, altering the balance
between consumers and resources, rendering ecosystems vulnerable to major disturbances (1, 2, 6, 7).

School of Plant Biology and Oceans Institute, The University

of Western Australia, 39 Fairway, Crawley, Western Australia
6009, Australia. 2Department of Environment and Agriculture,
School of Science, Curtin University, Bentley, Western Australia
6102, Australia. 3Department of Global Change Research,
Institut Mediterrani dEstudis Avanats (Universitat de les Illes
BalearsConsejo Superior de Investigaciones Cientficas),
Esporles, Spain. 4Commonwealth Scientific and Industrial
Research Organisation (CSIRO) Oceans and Atmosphere,
General Post Office Box 2583, Brisbane, Queensland 4001,
Australia. 5Service du Patrimoine Naturel, Musum National
d'Histoire Naturelle, 36 Rue Geoffroy Saint-Hilaire CP41, Paris
75005, France. 6Australian Institute of Marine Science, 39
Fairway, Crawley, Western Australia 6009, Australia. 7CSIRO
Oceans and Atmosphere Flagship, Private Bag 5, Wembley,
Western Australia 6913, Australia. 8Western Australian Museum,
Locked Bag 49, Welshpool DC, Western Australia 6986,
Australia. 9Research School of Biology, The Australian National
University, Canberra, Australian Capital Territory 2601,
Australia. 10Marine Science Program, Science Division,
Department of Parks and Wildlife, Kensington, Western
Australia 6151, Australia. 11School of Geography and
Environmental Science, The University of Western Australia, 39
Fairway, Crawley, Western Australia 6009, Australia. 12Marine
Biological Association of the United Kingdom, The Laboratory,
Citadel Hill, Plymouth PL1 2PB, UK. 13Marine Ecology Group,
School of Biological Sciences, The University of Canterbury,
Private Bag 4800, Christchurch, New Zealand. 14Instituto
Universitario Ecoaqua, Universidad de Las Palmas de Gran
Canaria, 35017 Canary Islands, Spain.
*Corresponding author. Email:
These authors contributed equally to this work. The other authors
are arranged alphabetically.

Now, climate change is also contributing to the erosion of resilience (8, 9), because increasing temperatures are modifying key physiological, demographic,
and community-scale processes (8, 10), driving species redistribution at a global scale and rapidly
breaking down long-standing biogeographic boundaries (11, 12). These processes culminate in novel ecosystems where tropical and temperate species
interact, with unknown implications (13). Here we
document how a marine heat wave caused the loss
of kelp forests across ~2300 km2 of Australias
Great Southern Reef, forcing a regime shift to
seaweed turfs. We describe a rapid 100-km range
contraction of kelp forests and a community-wide
shift toward warm-water species, with ecological
processes suppressing kelp forest recovery.
To document ecosystem changes, we surveyed
kelp forests, seaweeds, fish, mobile invertebrates,
and corals at 65 reefs across a ~2000-km tropicalto-temperate transition zone in western Australia
(14). Surveys were conducted between 2001 and
2015, covering the years before and after an extreme marine heat wave affected the region.
The Indian Ocean adjacent to western Australia
is a hot spot where the rate of ocean warming is
in the top 10% globally (15), and isotherms are
shifting poleward at a rate of 20 to 50 km per
decade (16). Until recently, kelp forests were dominant along >800 km of the west coast (8), covering
2266 km2 of rocky reefs between 0 and 30 m depth
south of 27.7S (Fig. 1). Kelp forests along the midwest section of this coast (27.7 to 30.3S) have
experienced steadily increasing ocean temperatures since the 1970s, recently punctuated by three
of the warmest summers in the past 215 years
(Fig. 2) (17, 18, 19). In December 2010, immediately
before an extreme marine heat wave, kelp forests
covered over ~70% of shallow rocky reefs in the
midwest (Fig. 2 and fig. S1), with no differences
in kelp cover or biomass among reefs along the
west coast (Fig. 1 and figs. S1 and S2) (8). During
this time, seaweed and fish communities in the
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


midwest were similar to those of the temperate

southwest (~500 km farther south) and clearly
distinct from those of tropical reefs in the (~500 km
farther north) (Fig. 3, A and B, and fig. S3) (17).
By early 2013, only 2 years later, our extensive
surveys found a 43% (963 km2) loss of kelp forests
on the west coast (Fig. 1). Previously dense kelp
forests north of 29S had disappeared (Fig. 2 and
fig. S1) or been severely reduced (>90% loss, Fig. 1),
representing a ~100-km range contraction and
functional extinction from 370 km2 of reef (a
reduction in abundance severe enough to delete
ecological function). Instead, we found a dramatic
increase in the cover of turf-forming seaweeds
(Fig. 2) and a community-wide shift from species
characteristic of temperate waters to species and
functional groups characteristic of subtropical and
tropical waters [Figs. 3 and 4 and fig. S3, mean
square contingency coefficient (2,52) = 0.70, P <
0.001]. Compared to the composition of the heavily
affected midwest reef communities in Kalbarri
before the 2011 marine heat wave, differences in
community structure (Bray-Curtis dissimilarity)
from reefs in Perth in the temperate southwest
increased by 91 and 28% for seaweeds and fishes,
respectively, whereas differences from Ningaloo
Reef in the tropical northwest decreased by 32
and 16%, respectively. This broad-scale communitywide reef transformation reflected consistent decreases in the abundance of taxa characteristic of
temperate reefs, coinciding with increases in the
abundance of species characteristic of subtropical
and tropical reefs, for both seaweeds (Fig. 3C and
table S2, 2,20 = 0.81, P < 0.001) and fishes (Fig.
3D and table S3, 2,20 = 0.64, P = 0.008). Similar
changes were seen for sessile and mobile invertebrates in the southern part of the midwest
region, where small hermatypic coral colonies increased almost sixfold in abundance and doubled
in species richness (Fig. 4 and table S5), while
the abundances of sea urchins and gastropods
also increased and decreased in accordance with
their thermal affinities (Fig. 4 and tables S4 and
S5, 2,12 = 0.68, P < 0.045).
Even though the acute climate stressor has
abated (Fig. 2 and fig. S1), as of late 2015, almost
5 years after the heat wave, we have observed no
signs of kelp forest recovery on the heavily affected
reefs north of 29S. Instead, concurrent with an
80% reduction in standing seaweed biomass (fig.
S2), we have recorded subtropical and tropical
fish feeding rates on canopy seaweeds that are
three times higher than on comparable coral reef
systems. Similarly, we have found a 400% increase
in the biomass of scraping and grazing fishes, a
functional group characteristic of coral reefs, which
now display grazing rates on seaweed turfs that
are comparable to those observed on healthy coral
reefs worldwide (table S6) (10). High herbivore
pressure now suppresses the recovery of kelp
forests by cropping turfs and kelp recruits (10).
We deduce that extreme temperatures beginning in 2011 exceeded a physiological tipping point
for kelp forests north of 29S, and now reinforcing
feedback mechanisms have become established
that support a new kelp-free state. Similar ecosystem
changes have not been observed in the south170

8 JULY 2016 VOL 353 ISSUE 6295

Fig. 1. Extent of kelp

forests in western
Australia before and after
the 2011 marine heat
wave. (Left) The extent of
kelp forests from 0 to 30 m
depth before 2011 is shown
(map), with a color scale
indicating the proportion
lost by 2013. (Right) The
area of kelp forests before
2011 in 0.5 bins (bars).
Before 2011, kelp forests
covered 2266 km2 along
the >800 km of coastline.
However, by 2013, 43% of
these kelp forests had
disappeared (red bars).
On the left side of the map,
gray squares (southwest
region) and black circles
(midwest region) mark
locations where reefs were
surveyed by scuba divers
to establish proportional
kelp loss (table S1).

Fig. 2. Regime shift from kelp forests to seaweed turfs after the 2011 marine heat wave. Kelp
forests were dense in Kalbarri until 2011 (A), when they disappeared from ~100 km of coastline (Fig. 1) and
were replaced by seaweed turfs (B). (C) The habitat transition (lines) coincided with exceptionally warm
summers in 2011, 2012, and 2013 (red bars), punctuating gradually increasing mean ocean temperatures
over the past decades (17). Shown are mean ( SE, n = 3 reefs) kelp forest cover (dark blue circles and line) and
seaweed turf cover (dark red circles and line), chronologically aligned with monthly sea surface temperature
(SST) anomalies (blue and red bars) relative to monthly climatological means for 19812015 (table S1).

west, where heat wave temperatures remained

within the thermal tolerance of kelps (17), and
the greater distance to tropical bioregions limited
the incursion of tropical species. Threshold temperatures for kelp forests appear close to 2.5C above
long-term summer maximum temperatures, consistent with other seaweeds in the region (20).
However, the partial loss of kelp forests on reefs

between 29 and 32S suggests that there is

variation in threshold temperatures within and
between kelp populations.
The consistent responses of both cool- and
warm-water species clearly illustrate the important
role of temperature. However, the transition in
community structure and subsequent persistence
of the new regime would have been augmented SCIENCE

Downloaded from on July 8, 2016


Fig. 3. Changes in seaweed and fish communities on affected reefs after the 2011 marine heat
wave. Ordinations (nonmetric multidimensional scaling) of seaweed (A) and fish (B) communities
show how community structure on reefs north of 29S shifted from a close resemblance to temperate reefs
farther south to a greater resemblance to tropical reefs to the north. Dark blue = Perth (20052007), light
blue = Kalbarri before (20052007), pink = Kalbarri after (20132015), red = Ningaloo Reef (2010) (table
S1). Each symbol represents an individual reef. Ordinations are based on Bray-Curtis dissimilarities calculated
from ln(x + 1)transformed data. (C and D) Change in ln(x + 1)transformed abundance of seaweeds [(C) grams
of fresh weight per 1.5 m2] and fish [(D) individuals per 2500 m2] in Kalbarri (20052007 versus 20132015)
clearly show the decline in cool-water species (blue bars) and concurrent increase in warm-water species (red bars),
with several species not previously recorded (+) or now absent from the samples (). Each bar represents an
average across six reefs for an individual species. White bars indicate taxa with ambiguous distributions. Species
are listed in tables S2 and S3.

Fig. 4. Changes in benthic invertebrate abundances in the midwest after the 2011 marine heat
wave. (A and B) Change in ln(x + 1)transformed abundance of common mobile invertebrates (inverts)
[(A) individuals per 30 m2] and small (<6 cm) hermatypic corals [(B) colonies per 1000 m2]. Colors and
symbols are as in Fig. 3. Each bar represents an average across 6 and 23 reefs for an individual species
of mobile invertebrates and corals, respectively. Mobile invertebrates were counted in Jurien Bay (2005,
2011 versus 2013 and 2014), and corals were counted between Cervantes and Dongara (30.6 to 29.3S)
(20052006 versus 2013) (table S1). Species are listed in tables S4 and S5.


by the low and high availability of temperate and

tropical propagules and immigrants, respectively,
as well as changes in competitive interactions
after the loss of kelp canopies (21). The oceanography of the region is dominated by the polewardflowing Leeuwin Current, which delivers warm
nutrient-poor water and tropical species into the
temperate region, while limiting the supply of
propagules, including kelp zoospores from higherlatitude kelp forests (22, 23). Indeed, healthy coral
reefs already occur at the Houtman-Abrolhos
Islands, 60 km offshore from Kalbarri, directly in
the path of the Leeuwin Current. Until now, however, cooler coastal waters have enabled kelp forests
to dominate nearshore reefs.
The Leeuwin Current is strongly influenced by
the El NioSouthern Oscillation (24). The strength
of the Leeuwin Current and the flow of warm
tropical water down the west coast of Australia
increase during the La Nia phase of this cycle.
These La Nia conditions drive warming anomalies
such as the 2011 marine heat wave (24) and are
predicted to double in frequency and intensity
in the near future (25). Moreover, the southeast
Indian Ocean has gradually warmed by at least
0.65C over the past five decades and will continue to warm until the end of the century and
beyond (19). Kelp forest recovery from disturbances
can be slow in the nutrient-poor west coast waters,
even in the cool southwest (8, 26), providing time
for populations of herbivorous fish to become
established and seaweed turfs to proliferate. Consequently, the probability of prolonged cool conditions that could reset community structure and
ecological processes to facilitate the recovery of
kelp forests is becoming increasingly unlikely,
while the risk of more heat waves that will exacerbate and expand the new tropicalized ecosystem state is increasing (25).
Short-term climate variability has previously
precipitated large-scale destruction of kelp forests
(27, 28), which have mostly recovered as environmental conditions returned to normal. The future
of kelp forest communities in western Australia
is, however, grim. Warming, more frequent heat
waves, and the intrusion of tropical species into
temperate habitats are unequivocal (13, 19, 25). The
current velocity of ocean warming is pushing kelp
forests toward the southern edge of the Australian
continent (29), where they are at risk of rapid local
extinction over thousands of kilometers, due to
the east-west orientation of the continents poleward coastline and west-to-east flow of surface
currents (12). This would devastate lucrative
fishing and tourism industries worth more than
$10 billion (Australian) per year (30) and have
catastrophic consequences for the thousands
of endemic species (30) supported by the kelp
forests of Australias Great Southern Reef.

T. P. Hughes, Science 265, 15471551 (1994).

R. S. Steneck et al., Environ. Conserv. 29, 436459 (2002).
M. S. Thomsen et al., Integr. Comp. Biol. 50, 158175 (2010).
M. Scheffer, S. Carpenter, J. A. Foley, C. Folke, B. Walker,
Nature 413, 591596 (2001).
5. N. A. J. Graham et al., Front. Ecol. Environ 11, 541548 (2013).
6. R. S. Steneck, A. Leland, D. C. McNaught, J. Vavrinec, Bull. Mar.
Sci. 89, 3155 (2013).

8 JULY 2016 VOL 353 ISSUE 6295

17 1

Downloaded from on July 8, 2016



21. D. P. Thomson, R. C. Babcock, M. A. Vanderklift, G. Symonds,

J. R. Gunson, Estuar. Coast. Shelf Sci. 96, 105113 (2012).
22. M. Feng, D. Slawinski, L. E. Beckley, J. K. Keesing, Mar. Freshw.
Res. 61, 12591267 (2010).
23. M. A. Coleman et al., J. Ecol. 99, 10261032 (2011).
24. M. Feng, M. J. McPhaden, S.-P. Xie, J. Hafner, Scientific Reports
3, 1277 (2013).
25. W. Cai et al., Nat. Clim. Change 5, 132137 (2015).
26. B. D. Toohey, G. A. Kendrick, E. S. Harvey, Oikos 116,
16181630 (2007).
27. P. K. Dayton, M. J. Tegner, Science 224, 283285 (1984).
28. E. A. Martinez, L. Cardenas, R. Pinto, J. Phycol. 39, 504508 (2003).
29. M. T. Burrows et al., Nature 507, 492495 (2014).
30. S. Bennett et al., Mar. Freshw. Res. 67, 4756 (2016).

This work was funded by the Australian Research Council (T.W.,

G.A.K.), the Hermon Slade Foundation (T.W., S.B.), a U.K. Natural
Environment Research Council Independent Research Fellowship
(D.A.S.), the Australian Institute of Marine Science (T.W., M.D.,
B.R.), the Australian National University (C.J.F.), the Western
Australian Museum (J.F.), the Department of Parks and Wildlife
(T.H.H., S.W.), CSIRO Oceans and Atmosphere (R.C.B., F.D.,


Structural basis for membrane

anchoring of HIV-1 envelope spike
Jyoti Dev,1,2* Donghyun Park,3* Qingshan Fu,1 Jia Chen,3,4 Heather Jiwon Ha,3,4
Fadi Ghantous,5 Tobias Herrmann,2 Weiting Chang,3 Zhijun Liu,6 Gary Frey,3,4
Michael S. Seaman,5 Bing Chen,3,4 James J. Chou1,6
HIV-1 envelope spike (Env) is a type I membrane protein that mediates viral entry.
We used nuclear magnetic resonance to determine an atomic structure of the transmembrane
(TM) domain of HIV-1 Env reconstituted in bicelles that mimic a lipid bilayer. The TM forms a
well-ordered trimer that protects a conserved membrane-embedded arginine. An aminoterminal coiled-coil and a carboxyl-terminal hydrophilic core stabilize the trimer. Individual
mutations of conserved residues did not disrupt the TM trimer and minimally affected
membrane fusion and infectivity. Major changes in the hydrophilic core, however, altered the
antibody sensitivity of Env. These results show how a TM domain anchors, stabilizes, and
modulates a viral envelope spike and suggest that its influence on Env conformation is an
important consideration for HIV-1 immunogen design.

IV-1 envelope spike [Env; trimeric (gp160)3,

cleaved to (gp120/gp41)3] fuses viral and
host cell membranes to initiate infection
(1). Binding of gp120 to receptor (CD4) and
co-receptor (e.g., CCR5 or CXCR4) trigger
large conformational changes, leading to a cascade of refolding events in gp41 and ultimately
to membrane fusion (24). Mature Env spikes,

Department of Biological Chemistry and Molecular

Pharmacology, Harvard Medical School, 250 Longwood
Avenue, Boston, MA 02115, USA. 2Virology Program, Harvard
Medical School, 260 Longwood Avenue, Boston, MA 02115,
USA. 3Division of Molecular Medicine, Boston Childrens
Hospital, 3 Blackfan Street, Boston, MA 02115, USA.
Department of Pediatrics, Harvard Medical School, 300
Longwood Avenue, Boston, MA 02115, USA. 5Center for
Virology and Vaccine Research, Beth Israel Deaconess
Medical Center, 330 Brookline Avenue, Boston, MA 02215,
USA. 6State Key Laboratory of Molecular Biology, National
Center for Protein Science Shanghai, Shanghai Institute of
Biochemistry and Cell Biology, Chinese Academy of
Sciences, Shanghai 200031, China.
*These authors contributed equally to this work. Corresponding
author. Email:


8 JULY 2016 VOL 353 ISSUE 6295

(gp120/gp41)3, are the sole antigens on the virion

surface; they often induce strong antibody responses in infected individuals (5, 6). A vast amount
of structural information is available for the ectodomain of Env, a primary target of the host immune system, but much less for its transmembrane
domain (TMD), membrane-proximal external region (MPER), and cytoplasmic tail (CT), in the
context of lipid bilayer. The cryo-EM (electron
microscopy) structure of a detergent-solubilized
clade B JR-FL EnvDCT construct without the CT
has been described recently (7), but its MPER and
TMD are disordered, probably because detergent
micelles failed to mimic a membrane environment.
The HIV-1 TMD is more conserved than a typical membrane anchor (fig. S1). Previous studies
showed that mutations and truncations in the
TMD indeed affect membrane fusion and viral
infectivity (811). Presence of a GxxxG motif, often
implicated in oligomeric assembly of TM helices
(12), suggests clustering of TMDs in membrane
(fig. S1). The presence of a conserved, positively

M.A.V.), Fisheries Research and Development Corporation project

no. 2008/013 (R.K.H., G.A.K.), The Marsden Fund of The Royal
Society of New Zealand (M.S.T.), and the WA Strategic Research
Fund for the Marine Environment (R.B., M.A.V., J.F.). T.W. and
S.B. conceptualized and wrote the manuscript; T.W., S.B., R.B.,
T.dB., K.C., M.D., F.D., J.F., C.J.F., J.S.-G., R.K.H., E.S.H.,
T.H.H., G.K., B.R., B.J.S., D.K.S., M.T., C.T., F.T., M.A.V., and S.W.
provided data; and T.W., S.B., R.K.H., J.S.-G., and D.A.S. performed
analyses and modeling. All authors discussed the results and
commented on the manuscript. The data are provided in the
supplementary materials. Additional information can be obtained
from T.W. All authors declare no conflicting interests.
Materials and Methods
Figs. S1 to S3
Tables S1 to S6
References (3170)
18 January 2016; accepted 31 May 2016

charged residue (usually arginine) near the middle of the TMD suggests functions other than just
spanning a bilayer. TM helices of many cell surface
receptors are not merely inert anchors but play
essential roles in receptor assembly and signal transmission. For example, we have shown that CT
truncation affects the antigenic surface of the ectodomain of HIV-1 Env on the opposite side of the
membrane (13). Thus, understanding the physical
coupling (conformation and/or dynamics) between
the CT and the ectodomain mediated by the TMD
may guide design of immunogens that mimic
native, functional Env and induce broadly neutralizing antibodies (bnAbs).
To characterize the TMD structure by nuclear
magnetic resonance (NMR), we used a fragment
of gp41 (residues 677 to 716; HXB2 numbering,
fig. S1), derived from a clade D HIV-1 isolate
92UG024.2 (14). This construct, gp41HIV1D(677-716),
contains a short stretch of MPER (residues 677 to
683); the TM segment (residues 684 to 705), defined by hydrophobicity; and a fragment previously assigned to the CT domain [residues 706
to 716, containing a tyrosine-based sorting motif
(15, 16)]. The gp41HIV1D(677-716) protein was purified
and reconstituted into bicelles (fig. S2, A and B)
(1719) with an expected lipid-bilayer diameter of
~44 (fig. S2C) (20, 21), thereby incorporating the
refolded gp41HIV1D(677-716) into a membrane-like environment. The bicelle-reconstituted gp41HIV1D(677-716)
migrated on SDSpolyacrylamide gel electrophoresis (SDS-PAGE) with a size close to that of a
trimer (theoretical molecular mass 14.1 kDa) (fig.
S2D), suggesting that the protein was trimeric
and resistant to SDS denaturation. The reconstituted gp41HIV1D(677-716) protein in bicelles generated
an NMR spectrum with excellent chemical-shift
dispersion (fig. S3A). The equivalent protein constructs from isolates 92BR025.9 (clade C) and
92RU131.16 (clade G) gave similar NMR spectra
(fig. S3, B and C), suggesting that the TMDs of most
HIV-1 Envs have similar structures when reconstituted in bicelles. We completed the NMR
structure of gp41HIV1D(677-716) using a previously described protocol (figs. S4 and S5) (22, 23). The final
ensemble of structures converged to a root mean SCIENCE

Downloaded from on July 8, 2016

7. S. D. Ling, C. R. Johnson, S. D. Frusher, K. R. Ridgway, Proc.

Natl. Acad. Sci. U.S.A. 106, 2234122345 (2009).
8. T. Wernberg et al., Ecol. Lett. 13, 685694 (2010).
9. N. A. J. Graham, S. Jennings, M. A. MacNeil, D. Mouillot,
S. K. Wilson, Nature 518, 9497 (2015).
10. S. Bennett, T. Wernberg, E. S. Harvey, J. Santana-Garcon,
B. J. Saunders, Ecol. Lett. 18, 714723 (2015).
11. E. S. Poloczanska et al., Nat. Clim. Change 3, 919925
12. T. Wernberg et al., Curr. Biol. 21, 18281832 (2011).
13. A. Vergs et al., Proc. Biol. Sci. 281, 20140846 (2014).
14. Materials and methods are available as supplementary
materials on Science Online.
15. A. Hobday, G. Pecl, Rev. Fish Biol. Fish. 24, 415425 (2014).
16. M. T. Burrows et al., Science 334, 652655 (2011).
17. T. Wernberg et al., Nat. Clim. Change 3, 7882 (2013).
18. J. Zinke et al., Nat. Commun. 5, 3607 (2014).
19. J. Lough, A. Sen Gupta, A. J. Hobday, in Report Card of Marine
Climate Change for Australia: Detailed Scientific Assessment,
E. S. Poloczanska, A. J. Hobday, A. J. Richardson, Eds. (Hobart,
Tasmania, 2012);
20. S. Bennett, T. Wernberg, B. Arackal Joy, T. de Bettignies,
A. H. Campbell, Nat. Commun. 6, 10280 (2015).

Fig. 1. NMR structure of the gp41HIV1D(677-716) trimer in bicelles. (A) Ribbon

representation of the lowest-energy structure from the calculated ensemble.
The sphere representation of the top view (lower right) shows that the trimer
has no ion-permeable holes. (B) The N-terminal half of the structure with hydrophobic residues (orange) arranged in the coiled-coil pattern (right panel). (C) The
C-terminal half of the structure showing an array of polar residues that form the

square deviation of 0.95 and 1.44 for backbone

and all heavy atoms, respectively (fig. S6 and
table S1).
gp41HIV1D(677-716) is a tightly assembled trimer
~54 long, with the conserved arginine (R696)
near its midpoint (Fig. 1A). It shows a packing
arrangement not seen in any other known TM
helix dimers or trimers: Its N- and C-terminal
halves have different modes of assembly, with an
intervening kink. The N-terminal region is a conventional three-chain coiled-coil formed by residues 686 to 696 (Fig. 1B), including the GxxxG
motif. The C-terminal half does not show classic
knobs-into-holes interactions, but instead is
held together by a network of polar contacts,
mainly involving R707 and Q710, at the trimer
interface of the kinked helical segments (residues 704 to 712) (Fig. 1C). We call this interface
the hydrophilic core.
R696, near the middle of each TM helix (Fig.
1D), produces three unbalanced charges at the
center of the membrane. This residue occupies
a d position in the heptad sequence (Fig. 1B).
Its Cb points toward the threefold axis of the
trimer, while the rest of the side chain bends
away from the axis, placing the guanidinium
group in a peripheral hydrophobic pocket formed

C-terminal hydrophilic core. The network of polar contacts is hypothesized to stabilize the trimer. (D) Enlarged middle region of the structure showing the intramembrane R696 and its surrounding hydrophobic residues, as well as the
backbone oxygen of L692. Single-letter abbreviations for the amino acid residues
are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L,
Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

by L692, L695, and I697 (Fig. 1D). The backbone

carbonyl of L692 may form a hydrogen bond with
one of the guanidinium NH2 group of R696. He
of R696 showed a water nuclear Overhauser effect (NOE) in a 15N-edited NOE spectrum (fig. S4),
indicating the presence of an adjacent structured
water. Thus, the guanidinium group, presumably
charged at pH 6.7 under the NMR conditions, is
partially neutralized by hydrogen bonding with
the electronegative backbone oxygen of L692 and
the water molecule. The polarizability of the hydrophobic pocket that surrounds the guanidinium may also lower its pKa (acid dissociation
constant) from its value in aqueous solution.
Although well accommodated in the TMD trimer,
the intramembrane R696 may modulate the stability of the helical trimer if the helices dissociate
at any stage in assembly or fusion. The 1H-15N
correlation spectrum of the gp41HIV1D(677-716)
trimer showed inhomogeneous peak linewidth,
with the N-terminal half near R696 having the
most severe peak broadening, consistent with
conformational fluctuation (fig. S7).
To confirm membrane partition of the TMD
trimer, we used a paramagnetic probe, Gd(DOTA)
(24), to measure solvent exposure of the four arginine residues in the gp41HIV1D(677-716) trimer.

These arginines are distributed at different

positions along the TM helices and thus serve as
four depth markers. We measured intensity decrease of the arginine He-Ne correlation peaks
at increasing concentrations of Gd(DOTA). The
most solvent-exposed R707 showed the highest
sensitivity to Gd(DOTA), whereas the most buried
R696 was the least sensitive (Fig. 2A). R683 and
R709 are near opposite lipid headgroup regions
in the structure. R709 showed a greater resonance
broadening than did R683, indicating that the
latter is more deeply buried. We placed the
gp41HIV1D(677-716) trimer in the lipid bilayer so
that the four arginine positions were consistent
with their respective sensitivity to Gd(DOTA)
(Fig. 2B). This placement, which is consistent
with the surface distribution of hydrophobic,
polar, and charged residues (Fig. 2C), places
R696 in a fully hydrophobic environment, slightly
closer to the cytoplasmic side of the membrane.
The MPER segment is in the headgroup region
of the outer leaflet. The C-terminal segment, previously assigned to the CT, is at the headgroupwater interface of the inner leaflet.
To assess the contribution of specific residues
to TMD stability, we generated 12 gp41HIV1D(677-716)
mutants with single or double mutations, mainly
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


Fig. 2. Partition of the gp41HIV1D(677-716) trimer in lipid bilayer. (A) Measurement of membrane immersion depth of the four arginines using the watersoluble and membrane-inaccessible paramagnetic probe Gd-DOTA. The 1H-15N
correlation peaks of the arginine He-Ne recorded in q = 0.5 bicelles (left) showed
reduced peak intensity with increasing Gd-DOTA concentration (right) due to paramagnetic relaxation enhancement (PRE). R683 is shown in green, R696 in blue,
R707 in magenta, and R709 in orange. Error bars represent spectra noise level
normalized against maximum reference peak intensity. (B) Placement of the
trimer structure in the presumed DMPC lipid bilayer, which is slightly thinner

in the trimer interface. To introduce large-scale

changes in the hydrophilic core, we also created
mutants D(705716) and G690L/D(705716) with
residues 705 to 716 deleted, and mutants 704713
and G690L/704713 with residues 704 to 713 mutated (Fig. 3 and table S2). We analyzed these
mutants by SDS-PAGE after reconstitution in
bicelles (fig. S8). Most simple mutations, including I686A, I693A, and R696N, did not disrupt
the trimer completely, but only shifted the band
to molecular mass positions lower than that of
the wild type, indicating only partial trimer destabilization (fig. S8). We observed a similar pattern for mutants 704713, G690L/704713, and
D705716. The only mutant that migrated as a
monomer was G690L/D705716, with the GxxxG
motif mutated and the entire hydrophilic-core
region truncated (fig. S8). Thus, both the coiledcoil and the hydrophilic core contribute to the
extraordinary stability of the TMD, and either
one of them is sufficient to prevent the trimer
from complete dissociation in bicelles.
To elucidate functional roles of the structural
elements in the TMD, we mutated each of them
in the intact Env spike and analyzed the effect on
Env biogenesis, membrane fusion, and viral infectivity. We generated 27 Env mutants, using the
sequence of a clade A isolate 92UG037.8, with
single, double, or triple mutations in the coiledcoil region, R696 and its protecting residues, the
kink, the hydrophilic core, or combinations of
these elements (table S2). We also produced mutants 704713 and G690L/704713. When transiently transfected in 293T cells, all mutants expressed
comparable levels of Env, with similar extents
of cleavage between gp120 and gp41, as well as
similar cell-surface levels (figs. S9 and S10). At a
high Env expression level, at which cell-cell fusion
was resistant to most neutralizing antibodies,

8 JULY 2016 VOL 353 ISSUE 6295

than the characterized thickness of DOPC bilayer (35), such that the positions
of the four arginines are in accordance with the PRE results in (A).The positions
of the arginine side chains are indicated by the He (spheres) in the ensemble of
15 low-energy structures. The color scheme is the same as in (A). (C) Surface
representation of the lowest-energy structure positioned in DMPC bilayer as in
(B) for showing the surface distribution of hydrophobic, polar, and charged
amino acids.The hydrophobic residues (yellow) include A, I, L, F, V, P, and G; the
polar residues (cyan) include Q, N, H, S, T, Y, C, M, and W; the charged residues
(blue) include K, R, D, and E.

the fusion activity of the TMD mutants was indistinguishable (93 to 109%) from that of the
wild type (fig. S11). When the expression level was
reduced to mimic the low surface density on
HIV-1 virions (25), several TMD mutants, including R696A, had significantly lower cell-cell fusion
activity than did the wild type (table S2 and fig.
S12). Likewise, when these mutations were introduced into pseudoviruses, there were no major
differences in Env incorporation and processing,
but limited changes in infectivity for most of them
(table S2 and figs. S13 and S14). R696A had almost full wild-type infectivity, whereas 704713
and G690L/704713 had substantially lessjust
the opposite of their effects on cell-cell fusion
(table S2 and fig. S14).
To determine whether mutations in the TMD
can influence the antigenic structure of the Env
ectodomain and whether they might affect its antibody sensitivity, we used the cell-cell fusion assay
to test inhibition of each Env mutant by a trimerspecific bnAb PG16 and by a nonneutralizing V3
antibody 3791 (table S3). Most mutants were essentially identical to the wild type in their sensitivity to either antibody; mutants I686A, Y712A,
P714N, G690L/Y712A, G690L/P714N, G690L/P714K,
and G690L/P714H exhibited detectable differences
(fig. S15). We observed the most pronounced differences for mutants 704713 and G690L/704713,
for which PG16 and 3791 switched phenotypes
the former became inactive and the latter inhibitory (fig. S15). All mutations that affect antibody
inhibition are located in either the coiled-coil or
the hydrophilic core. We infer that changes in
TMD stability influence the antigenic structure
of the ectodomain of the functional Env. When
tested in a pseudovirus-based neutralization assay
with bnAbs PG9 (trimer specific), 3BNC117 (CD4
binding site), 10-1074 (glycan- and V3-dependent),

10E8 (MPER), and 3791 (table S3), most mutants,

except for 704713 and G690/704713, were unchanged in their sensitivity to PG9, 3BNC117, 101074 and 3791, but most became more sensitive
to 10E8 (table S4). This result suggests that the
changes produced by all the mutations tested,
except for 704713 and G690/704713, are limited
to local structure. In contrast, the mutants 704
713 and G690/704713 became resistant to PG9
and sensitive to 3791reflecting their properties
in the cell-cell fusion assay. Mutant 704713 was
analyzed with additional antibodies. For cell-cell
fusion, wild-type Env is sensitive to trimer-specific
bnAbs PG9, PG16, and PGT145 and resistant to
nonneutralizing antibodies b6 (CD4 binding site),
3791, and 17b (CD4-induced) (table S3, Fig. 3A,
and fig. S16). The antibody inhibition pattern is
reversed, however, for the fully functional mutant 704713, indicating that the hydrophilic
core of the TMD plays an important role in stabilizing and modulating the antigenic structure
of the Env spike. Similar phenotypes were also
observed with a 704713 mutant derived from a
clade C strain C97ZA012 (fig. S17). The pseudovirus neutralization assay gave similar, but less
pronounced, results for the mutant 704713 (Fig.
3B and table S4).
The most important finding from this study relevant to vaccine development is how the TMD
modulates the antigenic surfaces of the Env spike.
We reported previously that truncation of the CT
domain of HIV-1 Env reshapes the antigenic surfaces of its ectodomain (13). We now show that
mutations destabilizing the hydrophilic core of
the TMD trimer resemble the CT deletion in altering the sensitivity of the functional Env to
both nonneutralizing and trimer-specific neutralizing antibodies. In particular, the trimer-specific
bnAbs, which neutralize by stabilizing the native SCIENCE

Downloaded from on July 8, 2016



conformation of Env (26, 27), do not recognize the

Env spike when its TMD has been destabilized.
We suggest that the TMD mediates conformational coupling between the ectodomain and the
CT and that the trimeric structure seen by NMR
represents the conformation of the TMD adopted
by a native Env spike in a membrane.
The placement of gp41HIV1D(677-716) in a lipid
bilayer reveals clear boundaries of the TM segment and settles a contentious issue (10, 28). Part
of the 10E8 epitope (residues 677 to 683) is embedded in the headgroup layer of the outer leaflet, consistent with lack of accessibility of this
epitope on the native Env (13, 29). The hydrophilic core, which was thought to be part of the
CT, is similarly protected by the headgroup layer
of the inner leaflet. This hydrophilic region contains a tyrosine-based sorting signal (712YSPL715),
which may participate in Env internalization by
endocytosis (15, 16, 30). Our structure indicates
that Y712 and P714 in this motif on one TM
protomer interact with L704 and V708 of the
adjacent protomer, respectively, thereby also
contributing to trimer stability.
The TMD is required not only for membrane
anchoring and fusion, but also for stability of the
entire Env spike. This observation can explain
why most soluble Env preparations with the
TMD deleted, except for those of a few selected
strains (31, 32), are unstable and conformationally heterogeneous, unless they have specific,
stabilizing modifications (26, 33, 34). To design
immunogens that mimic optimally native viral
spikes, one must not ignore structural constraints
imposed by the TMD on the ectodomain. The
high-resolution structure of the HIV-1 Env TMD
trimer presented here can be a guide for engineering more effective immunogens.


1. S. C. Harrison, Nat. Struct. Mol. Biol. 15, 690698 (2008).

2. D. C. Chan, D. Fass, J. M. Berger, P. S. Kim, Cell 89, 263273 (1997).
3. W. Weissenhorn, A. Dessen, S. C. Harrison, J. J. Skehel,
D. C. Wiley, Nature 387, 426430 (1997).
4. M. Pancera et al., Nature 514, 455461 (2014).
5. X. Wei et al., Nature 422, 307312 (2003).
6. D. D. Richman, T. Wrin, S. J. Little, C. J. Petropoulos,
Proc. Natl. Acad. Sci. U.S.A. 100, 41444149 (2003).
7. J. H. Lee, G. Ozorowski, A. B. Ward, Science 351, 10431048 (2016).
8. R. J. Owens, C. Burke, J. K. Rose, J. Virol. 68, 570574 (1994).
9. L. Shang, L. Yue, E. Hunter, J. Virol. 82, 54175428 (2008).
10. L. Yue, L. Shang, E. Hunter, J. Virol. 83, 1158811598 (2009).
11. E. Helseth et al., J. Virol. 64, 63146318 (1990).
12. M. G. Teese, D. Langosch, Biochemistry 54, 51255135 (2015).
13. J. Chen et al., Science 349, 191195 (2015).
14. F. Gao et al., J. Virol. 70, 16511667 (1996).
15. J. F. Rowell, P. E. Stanhope, R. F. Siliciano, J. Immunol. 155,
473488 (1995).
16. M. Boge, S. Wyss, J. S. Bonifacino, M. Thali, J. Biol. Chem. 273,
1577315778 (1998).
17. M. E. Call et al., Cell 127, 355368 (2006).
18. M. E. Call, K. W. Wucherpfennig, J. J. Chou, Nat. Immunol. 11,
10231029 (2010).
19. To produce the protein, we expressed the His-tagged TrpLEgp41HIV1D(677-716) fusion protein in Escherichia coli as inclusion
bodies, purified the solubilized protein by Ni-affinity chromatography, removed the TrpLE tag with cyanogen bromide, and
separated the product by reverse-phase high-performance
liquid chromatography. Bicelles were made of 1,2-dimyristoylsn-glycero-3-phosphocholine (DMPC; lipid) and 1,2-dihexanoylsn-glycero-3-phosphocholine (DHPC; detergent) at a ratio (q)
of 0.5. In this report, we use gp41HIV1D(677716) and TMD
interchangeably for convenience.
20. K. J. Glover et al., Biophys. J. 81, 21632171 (2001).
21. C. R. Sanders 2nd, J. P. Schwonek, Biochemistry 31,
88988905 (1992).
22. B. OuYang et al., Nature 498, 521525 (2013).
23. For structure determination, we proceeded with the clade D
construct because its expression level was the highest. The
approach involves determination of local structures of the
monomers and assembly of the trimer with intermonomer distance
restraints derived from NOEs between structurally equivalent but
isotopically differently labeled subunits. We could identify eight
intermonomer NOEs using the isotopically mixed labeled sample to
calculate a unique assembly solution, which was further validated
and refined with other conventional NOE data.

24. Gd(DOTA) is a water-soluble and membrane-inaccessible molecule, so that the paramagnetic relaxation enhancement (PRE) it
generates decreases with distance from the bilayer surface.
25. P. Zhu et al., Nature 441, 847852 (2006).
26. R. W. Sanders et al., PLOS Pathog. 9, e1003618 (2013).
27. J. P. Julien et al., Proc. Natl. Acad. Sci. U.S.A. 110, 43514356 (2013).
28. J. T. West, P. B. Johnston, S. R. Dubay, E. Hunter, J. Virol. 75,
96019612 (2001).
29. J. Chen et al., J. Virol. 88, 12491258 (2014).
30. C. Berlioz-Torrent et al., J. Virol. 73, 13501361 (1999).
31. S. A. Jeffs et al., Vaccine 22, 10321046 (2004).
32. J. M. Kovacs et al., Proc. Natl. Acad. Sci. U.S.A. 111, 1854218547 (2014).
33. J. Guenaga et al., J. Virol. 90, 28062817 (2015).
34. Y. Do Kwon et al., Nat. Struct. Mol. Biol. 22, 522531 (2015).
35. M. C. Wiener, S. H. White, Biophys. J. 61, 434447 (1992).

We thank S. Harrison and S. Rits-Volloch for generous advice and

assistance; D. Barouch, B. Haynes, and A. Carfi for critical reading of the
manuscript; and the NIH AIDS Reagent Program, Division of AIDS,
National Institute of Allergy and Infectious Diseases, NIH, for reagents.
The data presented in this manuscript are tabulated in the main paper
and in the supplementary materials. The atomic structure coordinate
and structural constraints are deposited in the Protein Data Bank under
the accession number 5JYN. The chemical-shift values are deposited in
the Biological Magnetic Resonance Data Bank under the accession
number 30090. This work was supported by NIH grants AI084794 (to
B.C. and Dan H. Barouch), GM083680 (to B.C.), AI106488 (to B.C.),
HL103526 (to J.J.C.), and AI127193 (to B.C. and J.J.C.), Collaboration
for AIDS Vaccine Discovery (CAVD) grant OPP1040741 (to
Dan H. Barouch from the Bill and Melinda Gates Foundation), and
the Center for HIV/AIDS Vaccine ImmunologyImmunogen Design
AI-100645 (to Barton F. Haynes). The NMR data were collected at
the NMR facility of National Center for Protein Science Shanghai
(supported by Chinese Academy of Sciences grant XDB08030301)
and Massachusetts Institute of TechnologyHarvard Center for
Magnetic Resonance (supported by NIH grant P41 EB-002026).
Materials and Methods
Figs. S1 to S20
Tables S1 to S4
References (3652)
17 March 2016; accepted 13 June 2016
Published online 23 June 2016

8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016

relative fusion activity

% Neutralization

% Neutralization

relative fusion activity

Fig. 3. Effect of mutations in

mutant 704-713
92UG037.8 (wt)
the TMD of HIV-1 Env on its
antibody sensitivity. (A) Anti1
body inhibition of cell-cell
-- b6
-- b6
fusion mediated by the wild-- 3791
-- 3791
type 92UG037.8 Env (left) and
the TMD mutant 704713
-- PG9
-- PG9
-- PG16
-- PG16
[right; residues 704 to 713
-- PGT145
-- PGT145
(VINRVRQGYS) were mutated
-- VRC01
-- VRC01
to SSAASAAGSA] was ana0
lyzed with both nonneutralizing
antibody concentration (M)
antibody concentration (M)
antibodiesincluding b6 (CD4
binding site; blue), 3791 (V3;
92UG037.8 (wt)
mutant 704-713
cyan), and 17b (CD4-induced;
purple)and trimer-specific
--- b6
--- b6
bnAbs, including PG9 (orange),
--- 3791
--- 3791
PG16 (red), and PGT145
(magenta). The CD4 binding
--- PG9
--- PG9
site bnAb VRC01 (green) was a
--- PG16
--- PG16
control antibody. The experi--- PGT145
--- PGT145
ment was carried out in tripli--- 3BNC117
--- 3BNC117
cate and repeated at least
twice with similar results. Error
antibody concentration (g/ml)
antibody concentration (g/ml)
bars indicate the SD calculated
by the Excel STDEV function. (B) Antibody neutralization of pseudoviruses containing either the 92UG037.8 Env (left) or the TMD mutant 704713 (right)
was determined with antibodies b6, 3791, 17b, PG9, PG16, and PGT145, shown in the same color scheme as in (A). The CD4 binding site bnAb 3BNC117
was a control antibody (green). The experiment was performed in duplicate.


Cilia-based flow network in

the brain ventricles
Regina Faubel,1 Christian Westendorf,2 Eberhard Bodenschatz,2 Gregor Eichele1*
Cerebrospinal fluid conveys many physiologically important signaling factors through the
ventricular cavities of the brain. We investigated the transport of cerebrospinal fluid in the third
ventricle of the mouse brain and discovered a highly organized pattern of cilia modules, which
collectively give rise to a network of fluid flows that allows for precise transport within this
ventricle. We also discovered a cilia-based switch that reliably and periodically alters the flow
pattern so as to create a dynamic subdivision that may control substance distribution in the
third ventricle. Complex flow patterns were also present in the third ventricles of rats and pigs.
Our work suggests that ciliated epithelia can generate and maintain complex, spatiotemporally
regulated flow networks.

he ventricular system of the brain consists

of four interconnected, cerebrospinal fluid
(CSF)filled cavities, lined with ependyma
whose apical surface bears bundles of motile
cilia (1). CSF directional flow through the
ventricles is driven by continuous CSF secretion
into the ventricles and by orchestrated beating of
the cilia. Motile cilia are eyelash-shaped cellular
protrusions containing a microtubule-based axoneme that, by bending, confers motility. The direction of cilia beating depends on coupling of planar
cell polarity signaling with hydrodynamic forces
(2, 3). CSF flow toward the exit of each ventricle
affords an efficient washout of waste (4) such as
neurotoxins and protein aggregates. Additionally,
CSF flow delivers nutrients, signaling molecules,
microRNAs, and exosomes (512). How these CSF
components reach their target sites is an important question that prompted us to investigate the
CSF flow dynamics along the ventricular walls.
We conducted this analysis in mice in the ventral
part of the third ventricle (v3V; Fig. 1A) because
of its simple geometry and close juxtaposition to
potential targets, including the physiologically
important periventricular hypothalamic nuclei
that control circadian rhythms, hormone release,
thermoregulation, blood pressure, satiety, and
feeding (Fig. 1B).
Organotypic v3V wall explant cultures were
established, and cilia-generated flow was visualized
by adding 1-mm fluorescent beads. Global flow was
determined by particle tracking, and near-wall flows
were extracted, yielding a highly complex flow map
(Fig. 1C) that was consistent among animals (the
number of animals used is given in table S1). Flow
from the inflow duct fanned out after passing by
the anterior commissure. A first flow was directed
anteriorly toward the optic recess (Fig. 1C, flow 1).
A second stream of the fan moved ventrally (flow
2). A third stream advanced posterodorsally (flow 3)
and confronted, at Bregma level 0.9, an opposing
flow (flow 4). When 70-kDa fluorescein isothio1

Max Planck Institute for Biophysical Chemistry, Am

Fassberg 11, 37077 Gttingen, Germany. 2Max Planck
Institute for Dynamics and Self-Organization, Am Fassberg
17, 37077 Gttingen, Germany.

*Corresponding author. Email:


8 JULY 2016 VOL 353 ISSUE 6295

cyanate (FITC)dextran was applied locally to

the head of flow 6, it propagated as prescribed by
streamlines of the flow map (Fig. 1D). Movie S1
shows the flow of FITC-dextran when applied to
the heads of flows 1, 2, or 3. Injections to flows
1 and 2 resulted in an initially coherent flow that
then fanned out, thereby dispersing the tracer over
a sizable territory. Local application to the head of
flow 3 revealed the ability of ciliated ependyma to

produce a winding flow that replicates the shape

of the streamlines seen in the flow map. Thus, the
flow map may reflect the propagation of macromolecules in CSF.
To explore how the confrontation of flows 3
and 4 affects transport along the wall, beads were
applied to v3V whole mounts (movie S2). Flow 4
carried beads anteriorly; near the site of confrontation with flow 3, beads moved into two mirrorsymmetrical streams that intimately followed the
ventricular wall and transported the beads down
into the v3V. This flow dynamic created a separatrix
that, together with flow 5, divided the v3V into two
volumes. This flow-induced boundary hindered
the passage of tracer. Flow 6 formed another intraventricular boundary that hindered flow 7 from
drawing tracer away from flow 4. The flow pattern
remained unchanged up to a distance of ~120 mm
above the ependyma (fig. S1), and flows along the
two walls were related by mirror symmetry (fig. S2).
The velocity map showed that flow velocity at
21C varied across the v3V wall within the range
of 150 to 500 mm/s (fig. S3, A and B). Thus, it takes
only a few seconds to move substances from the
entrance of the v3V to its floor. These velocities
are typical for cilia-driven flows (13).
We next analyzed cilia movement in the v3V
and found that the flow map matched the cilia
beating pattern. To show this, we captured movies

Fig. 1. Anatomy of the third ventricle in the mouse and flow map of the v3V. (A) Sagittal section
through the third ventricle (blue), consisting of a dorsal (d3V) and a ventral (v3V) part connected by two
ducts. As indicated by white arrows, CSF enters the v3V through the foramen interventriculare and exits
through the aqueduct. (B) The position of periventricular hypothalamic nuclei [reconstructed from (22)].
Blue and red outlines differentiate the d3V and v3V, and hatching indicates the choroid plexus (blue) and
median eminence (red). Medial preoptic nucleus, yellow; suprachiasmatic nucleus, gray; anterior hypothalamic
area, brown; paraventricular nucleus, pink; ventromedial nucleus, green; arcuate nucleus, orange; dorsomedial
nucleus, blue. (C) The flow map of the v3V, generated by particle tracking, shows that near-wall flow is subdivided into multiple flow domains. The associated eight major flow directions are indicated with turquoise
arrows. (D) FITC-dextran, applied at the head of flow 6 (asterisk) from a femtopipette (p), follows the streamlines of the flow map shown in the first panel (t, time). Scales in (B) to (D) are Bregma levels in millimeters.
a, anterior; ac, anterior commissure; aq, aqueduct; cx, cortex; d, dorsal; fiv, foramen interventriculare; mb,
midbrain; my, myelencephalon; p, posterior; S, separatrix; th, thalamus; tz, tanycyte zone; v, ventral. SCIENCE

Downloaded from on July 8, 2016


of rectangular fields of beating cilia bundles across

the entire v3V in a mosaic-like fashion. Using
custom-made software, the cilia movement was

extracted from each movie; results were then

stitched together to generate a v3V-wide map of the
time-averaged cilia beating pattern. At the separa-

Fig. 2. Cilia beating and fluid flow directions match. (A) Flow map of and (B) time-averaged cilia
beating orientation in the separatrix region [the flow map is color-coded as in Fig. 1C]. Computational
analysis of cilia movement reveals obliquely opposing cilia orientation (movie S4) that corresponds to the
pattern of the tracks shown in (A). (C) Flow map of two isolated fragments (I and II) of ependyma. (D and
E) Time series show the flow of locally applied FITC-dextran across fragments I and II, which were
recombined in vitro head to tail (D) or head to head (E). Red arrowheads flank the flow of FITC-dextran,
turquoise arrows indicate cilia beating direction, and yellow dashes denote fragment borders. Scale bars,
30 mm [(A) and (B)] and 100 mm [(C) to (E)].

Fig. 3. Temporal changes in flow.

(A) Flow map of the v3V at ZT23. A
whirl is circumscribed by the ellipse.
The scale shows Bregma levels in
millimeters. Solid and dashed arrows
indicate the direction of flow. (B) Timeaveraged cilia beating orientation in
the whirl, showing a counterclockwise
orientation (movie S6). Scale bar, 50 mm.
(C and D) Dynamics of FITC-dextran
propagation at ZT10 (C) and ZT23 (D),
as shown in movies S8 to S11, superimposed on the outline of the v3V. Injection points are marked by red asterisks.
Near-wall flow of low-dose injections is
shown with white dashes. Nested color
fields trace the dye propagation for highdose injections from a tip placed into
the field of most intense color. Gray
arrows represent flow directions taken
from Figs. 1C (C) and 3A (D). A black
dashed line represents the separatrix. The red circle in (D) represents a reference mark from movie S10.


trix, the flow pattern (Fig. 2A) corresponded to the

beating orientation of underlying cilia (Fig. 2B
and movie S4). At the entrance of the v3V, the tripartite flow also corresponded to the cilia beating
orientation (fig. S4 and movie S5). Thus, cilia form
distinct modules in which they beat coherently
and, in this way, locally control the flow of the
overlying fluid. As these two examples illustrate,
the arrangement of cilia modules accounts for the
complex flow map with its distinct flow domains.
When the v3V wall was cut into pieces, the flow
pattern in each fragment was virtually identical
to that observed at the intact wall before cutting
(fig. S5). Thus, cilia modules may determine flow
independently of context. In keeping with this
idea, we isolated two fragments (I and II) from
the v3V wall (Fig. 2C) and recombined them at
different angles. When the two fragments were
combined head to tail, focally applied FITC-dextran
translocated from one piece to the next (Fig. 2D).
When the same two fragments were oriented with
cilia beating in opposite directions, tracer applied
to fragment I progressed toward the interfragment boundary, at which the flow was diverted
upward by the counterflow originating from
fragment II (Fig. 2E). Thus, transport along bent
flows and formation of boundaries can be achieved
by appropriate arrangement of cilia modules.
Furthermore, cilia beating direction in each module is necessary and sufficient to sculpt the associated flow domains.
There are major physiological differences between an animal at rest (Zeitgeber time, ZT0 to
ZT12) and an awake animal (ZT12 to ZT24) (14).
The flow maps at ZT10 (Fig. 1C) and ZT23 (Fig. 3A)
showed differences. At ZT23, we observed a prominent whirl ventrally of the separatrix. This whirl
resulted from a circular arrangement of cilia
beating (Fig. 3B). Movie S6 shows how this counterclockwise whirl relates to the separatrix. Recording
the same region at ZT10 revealed a substantially
different pattern devoid of a whirl (movie S7). At
ZT10, separatrix and flow 5 formed a vertical intraventricular boundary. When a low dose of FITCdextran was injected into flow 4, the tracer flowed
ventrally, along the boundary, to reach the floor
(white dashes in Fig. 3C, movie S8). This boundary
withstood a high FITC-dextran dose (blue fields in
Fig. 3C, movie S9). When low or high doses of
FITC-dextran were injected into flows 3 or 4 at
ZT23, tracer remained in the dorsal part of the v3V
(Fig. 3D and movies S10 and S11), indicating the
presence of a dorsoventral boundary in this region.
The appearance of a whirl at ZT23 leads to a
major change in the global flow pattern: The vertical intraventricular boundary that was present
at the end of the resting phase was absent at the
end of the awake phase. This raises the possibility
that the distribution of CSF varies with the time
of the day. Our FITC-dextran tracer experiments
show that mixing of the flow is very weak. This is
also supported by comparing the time scale of
mixing, estimated from the flow gradient, with the
time that it takes the flow to transverse the ventricle.
Mammalian third ventricles differ in size and
morphology (15), yet both the modular organization and the flow domains are largely conserved
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 8, 2016


fate by coexpressing a combination of transcription factors (20). The later descendants of these
radial glia cells also form the ependymal cell
sheet (21), and thus cilia modules therein might
also be specified by a particular combination of
regionally expressed transcription factors. Such
orchestrated development would establish a direct relationship between potential targets, the
hypothalamic nuclei, and the structure of the
transport network.
Fluid dynamics has long described complex
types of movement in other systems. Now we show
that complex fluid movements are also present in
the brain in the form of a transport network driven
by coordinated cilia beating patterns. In addition
to targeted substance delivery, cilia generate separating flows and whirls that may establish and
modulate intraventricular boundaries capable
of concentrating locally released compounds and
preventing their entry into off-target regions. Transient local changes in the beating pattern evoked a
major change in ventricular subdivision. The cellular and structural bases that underly such changes
in beating direction are unknown but may be driven
by transient changes in cell-cell interactions and
in planar cell polarity.

Fig. 4. Flow maps of rat and pig v3Vs. (A) The flow map of the rat v3V strongly resembles that of the
mouse v3V, except for an anteroventral whirl (ellipse), an S-shaped flow (1), and a posterior flow (9) that
emerges from the floor of the posterior part of the v3V. (B) The flow map of the pig v3V shows that it is
highly organized and has cilia modules, separatrices, and whirls. Scale bars, 1 mm. Color coding is as in the
previous figures.

1. M. R. Del Bigio, Acta Neuropathol. 119, 5573 (2010).

2. B. Guirao et al., Nat. Cell Biol. 12, 341350 (2010).
3. E. R. Brooks, J. B. Wallingford, Curr. Biol. 24, R973R982
4. W. C. Worthington Jr., R. S. Cathcart 3rd, Ann. N. Y. Acad. Sci.
130, 944950 (1966).
5. A. Kramer et al., Science 294, 25112515 (2001).
6. M. Y. Cheng et al., Nature 417, 405410 (2002).
7. S. Kraves, C. J. Weitz, Nat. Neurosci. 9, 212219 (2006).
8. S. C. Robins et al., Nat. Commun. 4, 2049 (2013).
9. A. Guldbrandsen et al., Mol. Cell. Proteomics 13, 31523163
10. Y. Zhang et al., J. Proteomics 119, 9099 (2015).
11. D. Chiasserini et al., J. Proteomics 106, 191204 (2014).
12. K. Sawamoto et al., Science 311, 629632 (2006).
13. J. B. Freund, J. G. Goetz, K. L. Hill, J. Vermot, Development 139,
12291245 (2012).
14. C. Richter, I. G. Woods, A. F. Schier, Annu. Rev. Neurosci. 37,
503531 (2014).
15. R. Nieuwenhuys, H. J. ten Donkelaar, C. Nicholson, The Central
Nervous System of Vertebrates (Springer-Verlag, 1998).
16. B. Flix et al., Brain Res. Bull. 49, 1137 (1999).
17. T. C. Mathew, Anat. Histol. Embryol. 37, 918 (2008).
18. M. Bolborea, N. Dale, Trends Neurosci. 36, 91100 (2013).
19. M. Shimada, T. Nakamura, Exp. Neurol. 41, 163173 (1973).
20. J. L. Ferran, L. Puelles, J. L. Rubenstein, Front. Neuroanat. 9,
46 (2015).
21. N. Spassky et al., J. Neurosci. 25, 1018 (2005).
22. K. B. J. Franklin, G. Paxinos, The Mouse Brain in Stereotaxic
Coordinates (Academic Press, 1997).

between mice and rats (compare Fig. 1C with Fig.

4A). The ventricular recesses in the rat are more
prominent than in the mouse. Accordingly, when
we examined cilia-generated flow in the v3V of
rats, flow 1, directed at the supraoptic recess, was
S-shaped, and a new flow (flow 9) emerged from
the posterior floor of the v3V, to which the infundibular and inframammillary recesses are connected.
The v3V of pigs differs in size and architecture
from those of rodents (16), yet we also observed
flow domains in pig v3V (Fig. 4B).
A combination of straight and bent flows guides
molecules across several hundred microns. For

8 JULY 2016 VOL 353 ISSUE 6295

example, flow initiating at the entrance of the

v3V extends as far as the arcuate nucleus. This
flow may play a role in delivery or uptake of substances in this region, which is characterized by
an abundance of chemosensory tanycytes (17).
Tanycyte cell bodies are in contact with CSF,
and their processes extend several hundred micrometers deep into the brain parenchyma, thus acting
as bridges between the ventricle and neurons (18).
The similarity in design of the transport networks across species suggests a genetic regulation of network development. Hypothalamic nuclei
arise from radial glia cells (19) that acquire their

We thank A. Bae, A. Pumir, and C. Lo for valuable comments and

suggestions; C. Thaller for technical assistance; J. Schrder-Schetelig
for providing multirecorder software; and D. Hornung and T. Baig
for providing euthanized minipigs. This work was supported by the
Max Planck Society.
Materials and Methods
Figs. S1 to S6
Table S1
References (2325)
Movies S1 to S11
9 December 2015; accepted 10 June 2016
10.1126/science.aae0450 SCIENCE

Downloaded from on July 8, 2016



Reengineering chimeric antigen

receptor T cells for targeted therapy
of autoimmune disease
Christoph T. Ellebrecht,1 Vijay G. Bhoj,2 Arben Nace,1 Eun Jung Choi,1 Xuming Mao,1
Michael Jeffrey Cho,1 Giovanni Di Zenzo,3 Antonio Lanzavecchia,4 John T. Seykora,1
George Cotsarelis,1 Michael C. Milone,2* Aimee S. Payne1*
Ideally, therapy for autoimmune diseases should eliminate pathogenic autoimmune cells
while sparing protective immunity, but feasible strategies for such an approach have been
elusive. Here, we show that in the antibody-mediated autoimmune disease pemphigus
vulgaris (PV), autoantigen-based chimeric immunoreceptors can direct T cells to kill
autoreactive B lymphocytes through the specificity of the B cell receptor (BCR). We
engineered human T cells to express a chimeric autoantibody receptor (CAAR), consisting
of the PV autoantigen, desmoglein (Dsg) 3, fused to CD137-CD3z signaling domains.
Dsg3 CAAR-T cells exhibit specific cytotoxicity against cells expressing anti-Dsg3 BCRs
in vitro and expand, persist, and specifically eliminate Dsg3-specific B cells in vivo.
CAAR-T cells may provide an effective and universal strategy for specific targeting of
autoreactive B cells in antibody-mediated autoimmune disease.

emphigus vulgaris (PV) is a life-threatening

autoimmune blistering disease caused by
autoantibodies to the keratinocyte adhesion protein Dsg3 (1). CD20-targeted B cell
depletion results in short-term disease remission in 95% of pemphigus patients, but 81%
relapse and fatal infection may occur (2). After
CD20-targeted depletion, serum autoantibody
titers drop, indicating that short-lived plasmablasts are the source of autoantibodies in PV
and targeting of CD20+ memory B cell precursors
indirectly depletes autoantibody-secreting CD20
plasmablasts (3, 4). Relapsing pemphigus demonstrates the same anti-Dsg3 B cell clones observed
during active disease, whereas disease remission
is associated with disappearance of circulating
anti-Dsg3 B cells (5). Thus, targeted elimination of
anti-Dsg3 memory B cells should cure PV without the risks of general immunosuppression.
Recently, chimeric antigen receptor (CAR) technology has revolutionized cancer therapy. A
CD19-specific CAR, consisting of an extracellular
single-chain variable fragment (scFv) antibody
to CD19 fused to T cell cytoplasmic signaling
domains, activates T cell cytotoxicity upon contact with CD19+ B cells, causing specific and
permanent elimination of B cells and durable
remission of leukemia (614). In PV, pathogenic
memory B cells express anti-Dsg3 B cell receptors
(BCRs). We reasoned that by expressing Dsg3 as
the extracellular domain of a chimeric immuno1

Department of Dermatology, University of Pennsylvania,

Philadelphia, PA 19104, USA. 2Department of Pathology and
Laboratory Medicine, University of Pennsylvania,
Philadelphia, PA 19104, USA. 3Laboratory of Molecular and
Cellular Biology, Istituto Dermopatico dellImmacolata
(IDI-IRCCS), 00167 Rome, Italy. 4Institute for Research in
Biomedicine, 6500 Bellinzona, Switzerland.
*Corresponding author. Email:
(A.S.P.); (M.C.M.) These authors
contributed equally to this work.


receptor, cytotoxicity would become specific for

only those B cells bearing anti-Dsg3 BCRs, providing targeted therapy for PV without general
immunosuppression. Such a strategy would directly eliminate surface immunoglobulin (sIg)+
anti-Dsg3 memory B cells and indirectly eliminate sIg Dsg3-specific short-lived plasma cells
that produce the disease-causing antibodies. We
thus created a chimeric autoantibody receptor
(CAAR) (fig. S1A), with the autoantigen Dsg3 as
the CAAR extracellular domain, in order to engineer T cells to kill autoimmune B cells in PV.
Dsg3 consists of five extracellular cadherin
(EC) domains, with residues important for cell
adhesion residing in EC1 and EC2 (15). Autoantibodies to EC1, EC2, EC3, EC4, and EC5 occur
in 91, 71, 51, 19, and 12% of PV sera; no PV sera
target only the EC4 and/or EC5 domains (16).
Since T cell activation depends on the intermembrane distance of the immunologic synapse (17),
we reasoned that shorter conformational fragments of Dsg3 should enhance CAAR efficacy.
We therefore designed a panel of Dsg3 CAARs
for expression in primary human T cells (Fig. 1A),
using Dsg3 EC1-3/EC1-4/EC1-5 as the extracellular domain, fused to a dimerization-competent
CD8a transmembrane (18) and CD137-CD3z cytoplasmic signaling domains, which were used
successfully in CD19 CAR clinical trials (6, 7).
EC1-3/EC1-4 CAARs demonstrate robust surface
expression of mature, conformational Dsg3 in
primary human T cells, whereas EC1-5 CAAR
expression is more variable (fig. S1, B to D).
We first evaluated the ability of Dsg3 CAAR-T
cells (CAAR-Ts) to kill anti-Dsg3 B cells in vitro,
using antibody-secreting hybridomas that target EC1 (AK23), EC2 (AK19), and EC3-4 (AK18)
(19), or K562/Nalm-6 cells expressing pathogenic
F779/anti-EC1 or PVB28/anti-EC2 IgG cloned from
PV patients (20, 21) (fig. S2). All BCRs were ex-

pressed at a density comparable to human B cells

(fig. S3). Dsg3EC1-3/EC1-4 CAAR-Ts demonstrate
interferon-g (IFN-g) secretion and specific cytolysis against anti-EC1/EC2 but not control targets (Fig. 1, B and C, and fig. S4). Dsg3EC1-5
CAAR-Ts exhibit minimal cytolysis, and Dsg3EC1-3
CAAR-Ts do not lyse anti-EC3/4 targets. Thus,
the Dsg3EC1-4 CAAR demonstrates the best combination of potency and breadth, with specific
cytolysis of cells expressing anti-Dsg3 BCRs.
To investigate the mechanism of CAAR-T activation, we examined the molecular organization within CAAR-Ts upon binding sIg target by
total internal reflection fluorescence microscopy.
CAAR-Ts form immunologic synapses analogous
to T cell receptor (TCR)peptide/major histocompatibility complex (MHC) interaction (22, 23),
with actin reorganization and centripetal movement of CAAR-IgG clusters resulting in a central
supramolecular activation complex (SMAC)like
structure (fig. S5A and movie S1). The protein
tyrosine phosphatase CD45 is excluded from early
CAAR-IgG microclusters (fig. S5B and movie S2),
similar to findings with the anti-CD19 CAR (24)
and the native TCR-MHC synapse (25), suggestive of a kinetic segregation model for CAAR activation (26).
PV patients have serum anti-Dsg3 IgG that
might neutralize, or alternatively could help stimulate, Dsg3 CAAR-Ts. We therefore evaluated
Dsg3 CAAR-T cytotoxicity in the presence of soluble monoclonal IgG matching the targeted BCR
to maximize the neutralization capacity of the
soluble IgG. Soluble anti-Dsg3 IgG decreases but
still preserves compelling CAAR-T cytotoxicity
against AK19/PVB28, minimally affects AK23/
AK18, and potentiates cytotoxicity against F779
targets (Fig. 2, A and B). Because CAAR-Ts will
encounter polyclonal anti-Dsg3 IgG in PV patients,
we next tested Dsg3 CAAR-T cytotoxicity against
polyclonal targets in the presence of polyclonal
PV serum IgG. Dsg3EC1-3, and to a lesser extent
Dsg3EC1-4, CAAR-Ts retained efficient levels of
cytolysis against AK hybridomas (Fig. 2C), despite the presence of PV serum IgG as well as
secreted anti-Dsg3 IgG by these hybridomas.
Both Dsg3EC1-3 and EC1-4 CAAR-Ts effectively
killed Nalm6 cells expressing human anti-Dsg3
IgG in the presence of PV serum IgG (Fig. 2D).
Surface binding competition assays indicate that
antibodies that inhibit cytotoxicity (AK19/PVB28)
persist on the CAAR-T surface, which reduces the
number of accessible CAAR molecules to bind
target cells, while noninhibitory F779/AK23/AK18
are more rapidly replaced by competing antibodies,
with AK18 replacement occurring even at 4C,
suggesting low affinity (Fig. 2, E and F). Indeed,
surface plasmon resonance analysis indicates that
noninhibitory antibodies tend to have faster
off-rates and lower affinity (Fig. 2, G and H,
and fig. S6A). Furthermore, F779 and PV serum
IgG stimulate Dsg3EC1-4 CAAR-Ts, more so than
the Dsg3EC1-3 CAAR-Ts, to secrete low levels of
IFN-g and proliferate (fig. S6, B and C). Thus,
the effect of soluble antibodies to Dsg3 on CAAR-T
function and activation is a complex process that
is affected by affinity and binding kinetics of the
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 7, 2016


Cytoplasmic tail


CD8 CD137 CD3



desmoglein 3 CAAR



Target: AK23 (anti-EC1)



VRC01 neg ctl






CART19 pos ctl
NTD neg ctl





Target: K562 F779 (anti-EC1)

K562 PVB28 (anti-EC2)


SS1 neg ctl vs. mesothelin+

EC1-4 CAAR vs. K562 wt
SS1 neg ctl vs. K562 wt









E:T ratio

Targets for control CARs


specific lysis [%]


EC1-4 CAAR vs. F779 IgG+

EC1-4 CAAR vs. PVB28 IgG+


E:T ratio

E:T ratio

Target: control hybridoma






E:T ratio


specific lysis [%]




Target: Nalm6 PVB28 IgG (anti-EC2)

CART19 pos ctl
NTD neg ctl




specific lysis [%]

Target: Nalm6 F779 IgG (anti-EC1)

VRC01 neg ctl

E:T ratio


specific lysis [%]




E:T ratio

specific lysis [%]



Target: AK18 (anti-EC3)


specific lysis [%]

specific lysis [%]

CART19 neg ctl



Target: AK19 (anti-EC2)






K K2
F7 62 (
79 CD1
K 9)
m K5
N PV 2
m B28
N F77
m 9

desmoglein 3 CAAR
lentiviral vector

CART19 (target: K562 CD19+)

VRC01 (target: K562 YU-2 gp120)
CART19 (target: K562 wt)
VRC01 (target: K562 wt)





E:T ratio




E:T ratio

Fig. 1. Dsg3 CAAR-T cells demonstrate specific and potent cytotoxicity. (A) Schematic of native Dsg3, Dsg3 CAAR, and lentiviral vector for stable CAAR
expression. (B) Specific IFN-g production by CAAR-Ts as measured by ELISA from coculture supernatant after 24 hours. Cultures were set up in duplicate; mean
values are shown. Similar results were obtained in independent experiments from at least five different healthy Tcell donors. (C) 51Cr release after 4 hours of Tcell
target cell coculture at indicated effector to target (E:T) ratios to measure specific cytotoxicity by Dsg3 CAAR-Ts against distinct anti-Dsg3 target cells. Mean
values of triplicate cultures are shown. Similar results were obtained in independent experiments from at least five different healthy T cell donors.

antibody, as well as the relative position of the

epitope in the CAAR molecule, highlighted by
the different susceptibility of the EC1-3 and EC1-4
CAARs to blockade and activation. Collectively,
these data suggest that soluble anti-Dsg3 IgG
will not prevent and may in fact promote CAART efficacy and persistence due to CD137-mediated
costimulatory signals that have been shown to
enhance CAR-T activity (2730) and ameliorate
T cell exhaustion (31).
We tested the in vivo efficacy of Dsg3 CAARTs against AK23/AK19/AK18 target cells in a
PV mouse model (19). We selected the PV hybridoma model for in vivo proof of concept, because this model allows preclinical testing of
human Dsg3 CAAR-T efficacy against B cells
targeting distinct and defined Dsg3 epitopes.

8 JULY 2016 VOL 353 ISSUE 6295

Furthermore, the mice have polyclonal serum

anti-Dsg3 IgG (like PV patients) that might neutralize or eliminate CAAR-Ts, and disease burden
can be serially and objectively quantitated by bioluminescence imaging. NSG (NOD-scid-gamma)
mice were injected with a polyclonal mixture of
bioluminescent AK18/AK19/AK23 hybridomas,
followed by injection with either Dsg3 CAAR-Ts
or control CAR-Ts at day 5 (fig. S7A), at which
time serum antibody titers to Dsg3 were detectable by enzyme-linked immunosorbent assay
(ELISA) (Fig. 3A). Dsg3 CAAR-Ts robustly control all PV hybridomas, resulting in decreased
Dsg3 serum autoantibody titers (Fig. 3A), absence
of autoantibody binding (Fig. 3B) and blistering
in oral mucosa (Fig. 3C), and significantly delayed
or no hybridoma outgrowth (Fig. 3D). Dsg3EC1-4

CAAR-Ts showed equal efficacy when tested

against each individual hybridoma in vivo (fig. S8,
A to G). We further analyzed a subset of Dsg3EC14 CAAR-Ttreated mice with increasing bioluminescence at day 18. Increasing flux was due to
outgrowth of sIg hybridomas that comprised
~1% of injected cells (fig. S8, H to J). This mechanism of CAAR-T escape observed in mice is irrelevant
in patients because sIg long-lived plasma cells
do not significantly contribute to PV autoantibody
production. Dsg3 CAAR-T frequencies in both
escape and cured mice were comparable, indicating that lack of CAAR-T cell persistence or
loss of Dsg3 CAAR expression is not responsible
for CAAR escape in vivo (fig. S8K).
We further tested the in vivo efficacy of
Dsg3EC1-4 CAAR-Ts against human CD19+ Nalm-6 SCIENCE

Downloaded from on July 7, 2016

desmoglein 3
chimeric autoantibody receptor

IFN [n g /m l]

native desmoglein 3
(pemphigus vulgaris antigen)

specific lysis [%]

CAAR T cell



Fig. 2. Dsg3 CAAR-T cells maintain cytotoxicity in the presence of soluble

antibodies to Dsg3. (A to D) 51Cr release after 4 hours of T celltarget cell
coculture at an E:Tratio of 30:1 to measure specific cytotoxicity by Dsg3 CAAR-Ts
in the presence of [(A) and (B)] soluble, monoclonal antibodies to Dsg3 at
indicated concentrations or [(C) and (D)] soluble, polyclonal antibodies to Dsg3
at indicated Dsg3 ELISA indices. Mean values from triplicate cultures are shown.
Results are representative of two independent experiments with CAAR-T cells
from two different donors. (E) Experimental setup for pulse-chase antibody cell-

B cells expressing anti-Dsg3 BCRs (F779/PVB28)

cloned from PV patients. This model additionally allows comparison with anti-CD19 CAR-Ts
(CART19) that have proven clinical efficacy (6, 7).
Although CAAR-T treatment is delayed until the
exponential growth phase of the target cells, Dsg3
CAAR-Ts effectively eliminate anti-Dsg3 Nalm-6
B cells with efficacy comparable to CART19, measured by bioluminescence and flow cytometric
analysis of bone marrow/spleen (Fig. 3, E to G,
and fig. S7, B and C). Collectively, these models
show that Dsg3 CAAR-Ts eliminate anti-Dsg3
BCR+ but not BCR cells in vivo, indicating specific cytolysis, as opposed to bystander activation
and off-target cytolysis, even in the presence of
soluble anti-Dsg3 IgG. Notably, CAAR-Ts persist
in the presence of soluble autoantibodies to Dsg3
in vivo (fig. S8K), indicating that circulating auSCIENCE

surface binding assay. (F) Total surface IgG and retained biotinylated IgG were
detected by flow cytometry after 6 hours of culture with excess nonbiotinylated
IgG at 4C and 37C to assess the blocking property of soluble IgG toward each B
cell clone. Similar results were replicated in three independent experiments with
Dsg3EC1-4 CAAR-T cells from three different T cell donors. (G and H) Surface
plasmon resonance demonstrating higher off-rates (kd) and lower affinity (higher
KD) of nonblocking antibodies AK18, AK23, and F779. Values were calculated
from four different concentrations, as indicated in fig. S6A.

toantibodies do not cause Fc-mediated clearance

of CAAR-Ts. Furthermore, CAAR-Ts showed engraftment and persistence in the spleen 3 weeks
after CAAR-T transfer, even in the absence of
target cells (fig. S8L), suggesting that persistence of CAAR-Ts does not depend on target cell
We next addressed CAAR-T safety by evaluating off-target effects against keratinocytes, which
express desmocollins and desmogleins, the physiologic binding partners of Dsg3. The Dsg3CAAR
does not exhibit CD3z-mediated signaling in reporter cells after incubation with keratinocytes
(Fig. 4, A and B), in contrast to positive control antiDsg3/1 (Px44) CAR. Correspondingly, Dsg3EC1-4
CAAR-Ts do not lyse human keratinocytes in vitro,
whereas Px44 CAR-Ts show potent lysis (Fig. 4C).
To evaluate toxicity in vivo, we tested CAAR-Ts in

human skin-xenografted mice (Fig. 4D). Px44

CAR-Ts exhibit exuberant epidermal infiltration
by day 3 (Fig. 4, D and E). Despite similar T cell
density in the graft dermis, Dsg3EC1-4 CAAR-Ts
do not demonstrate enhanced skin toxicity compared with CART19, which has shown no skin
side effects in humans. Additionally, no off-target
toxicity against other tissues was observed in
Dsg3EC1-4 CAAR-Ttreated mice (fig. S9). Because
human Dsg3 rescues the loss of mouse Dsg3, indicating functional interaction between human Dsg3
and mouse desmosomal components (32), these
experiments indicate that Dsg3EC1-4 CAAR-Ts
cause no toxicity to desmosome-bearing tissues in
vivo. The absence of off-target toxicity to desmosomebearing tissues might be explained by several factors. The intermembrane distance in desmosomes
is 35 nm (15), greater than the 15-nm optimum
8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 7, 2016



day 4

control CAR

10 10

10 10

10 10

10 9

10 9

10 9

day 18





10 8
10 7
10 6

10 8
10 7
10 6



total flux [P/s]

control CAR

total flux [P/s]

Serum Dsg3 ELISA


total flux [P/s]






10 8
10 7

control CAR

10 6


control CAR

10 9
10 8
10 7
10 6
10 5




10 11


total flux [P/s]


total flux [P/s]

total flux [P/s]





10 9
10 8
10 7
10 6
10 5




10 9
10 8
10 7
10 6
10 5




day 5
day 12



10 10



radiance (p/s/cm2/sr)



anti-mouse IgG




10 11



10 3

control CAR

CD19+ GFP+
cells (/l)


10 2

Dsg3 CART19


10 1
10 0
10 -1
10 -2

m bo
ar ne
sp w
m bo
ar ne
sp w
m bo
ar ne
sp w

10 -3

Fig. 3. Dsg3 CAAR-T cells eliminate anti-Dsg3 target cells in vivo. (A)
Serum anti-Dsg3 ELISA was performed on days 5 and 14 to quantify IgG
production by hybridoma cells before and after CAAR-T treatment. Paired t
test, two-tailed; *P < 0.05; ***P < 0.001. (B) Direct immunofluorescence of
mucosa samples to detect IgG deposition after CAAR-T or control T cell treatment. Absence of IgG deposition was seen in all CAAR-Ttreated mice. (C) Histologic mucosal blister formation (i.e., acantholysis) was assessed after CAAR-T
treatment. None of six CAAR-Ttreated mice, versus five of six control Tcell
treated mice, demonstrated acantholysis; P = 0.015, Fishers exact test. Scale
bars in (B) and (C), 50 mm. (D) Serial quantification of hybridoma burden by

distance for T cell activation. Additionally, cadherincadherin affinity is in the micromolar range (33),
whereas CAR-T function typically requires nanomolar affinity (3436). Finally, Dsg3 truncation
(EC1-3/EC1-4) potentially compromises CAAR interaction with desmosomal ligands while preserving epitopes required for autoantibody binding,
thus preventing off-target toxicity.
We also considered that FcgR-expressing cells
bind serum antibodies to Dsg3 and might become targets for Dsg3 CAAR-Ts, although significant toxicity is unlikely because anti-Dsg3 IgG
comprises only a small fraction of total serum
IgG. In the presence of PV serum, cytotoxicity
against CD64+ (FcgR+) K562 cells was undetect182

8 JULY 2016 VOL 353 ISSUE 6295

bioluminescence imaging. Unpaired Mann-Whitney test, two-tailed, **P <

0.01; results were replicated in an independent experiment with Tcells from a
different donor. Symbols represent one mouse each; horizontal black lines
represent the mean of each group. (E) Bioluminescence imaging quantification of Nalm6 CD19+ B cells expressing PVB28/F779 IgG after Dsg3 CAAR-T,
anti-CD19 CAR-T (CART19), or nontransduced Tcell (NTD) treatment. (F) Bioluminescence on days 5 and 12 after Nalm6 injection. Dashed lines separate
images from different cages. (G) Flow cytometric quantification of Nalm6
cells in bone marrow and spleen in the three treatment groups 22 days after
Nalm6 injection.

able (fig. S10, A and B). Additionally, Dsg3EC1-4

CAAR-T treatment in the setting of circulating
anti-Dsg3 IgG does not reduce FcgR-expressing
cells such as neutrophils and monocytes in vivo
(fig. S10C), indicating that redirected killing via
CAAR-antibody-FcgR interaction does not occur.
Furthermore, CAAR-T reporter cells do not activate when exposed to bone marrow B cells that
harbor a polyreactive and/or autoreactive repertoire (37), in contrast to CART19, which reacts
with bone marrow B cells (fig. S11). Thus, toxicity
by Dsg3EC1-4 CAAR-Ts is undetectable, underscoring the specificity of the CAAR concept.
In summary, CAAR-Ts represent a targeted
approach to therapy of antibody-mediated auto-

immune diseases, with the potential for generation of long-term memory CAAR-Ts that can
potentially cure disease. CAAR-Ts expressing the
PV autoantigen Dsg3 specifically kill anti-Dsg3
target cells, even in the presence of circulating
autoantibodies, and without off-target toxicity. We
acknowledge that short-term observations of CAART safety and efficacy in preclinical models are
inherently limited. Nevertheless, several of the
limitations of CAR-T therapy for cancer likely will
not apply to CAAR-T therapy of autoimmunity.
Target cell escape due to epitope spreading or somatic mutation is impossible, because B cells that
no longer bind Dsg3 will become irrelevant to
disease. Additionally, memory B cells depend on SCIENCE

Downloaded from on July 7, 2016


10 11



EC1-4 CAAR, HaCat





primary human

F779 IgG K562





Px44 CAR: HaCat

F779 IgG K562

NTD target: HaCat

NTD target: K562 wt




specific lysis











E:T ratio



Px44 CAR


Px44 CAR



CD3+ cells/mm 2




Fig. 4. Dsg3 CAAR-Tcells do not show off-target toxicity. (A) Flow-cytometric

quantification of CAAR-mediated signal transduction, as indicated by nuclear
factor of activated Tcells (NFAT)driven green fluorescent protein (GFP) expression in Jurkat reporter cells, using anti-Dsg3/1 Px44 CAR as a positive control for
keratinocyte expression of Dsg3. Numbers indicate percentage of live, single cells
in the GFP-positive gate. (B) Mean fluorescence intensity of GFP expression from
(A). Representative data were replicated in at least five independent experiments
with keratinocytes from different human donors. (C) 51Cr release after T cell
target cell coculture at indicated E:T ratios to measure cytotoxicity by Dsg3

survival and activation signals through the BCR;

thus, anti-Dsg3 B cells that down-regulate sIg are
unlikely to persist or mature into antibody-secreting
cells. Tumor lysis and cytokine release syndrome
are also unlikely since Dsg3-reactive B cells are
predicted to comprise only a small fraction of the
total B cell population. Ultimately, various combinations of CAARs could be combined to maximize
efficacy in diseases in which autoantibodies targeting multiple autoantigens are pathogenic. Thus,
CAAR-T cells represent an innovative therapeutic
strategy that avoids the risks of general immunosuppression and can likely be applied to other
autoantibody-mediated diseases.

1. A. S. Payne, J. R. Stanley, in Dermatology in General Medicine,

K. Wolff et al., Eds. (McGraw Hill, New York, 2012), chap. 53.
2. N. Colliou et al., Sci. Transl. Med. 5, 175ra30


CAAR-Ts and controls against human HaCat keratinocytes. Mean values of

triplicate cultures are shown; similar results were obtained in two independent
experiments. (D) Microscopic analysis of human skin xenografts to evaluate
epidermal infiltration by Dsg3CAAR-Ts, positive control Px44 CAR-Ts, and negative control CART19. (Inset) Dyskeratotic keratinocytes surrounded by T cells.
Dashed line shows dermal-epidermal junction. Scale bar, 125 mm. (E) Quantification of T cell infiltration into epidermis and dermis of human skin xenografts.
Each symbol represents one mouse; ratio-paired t test, two-tailed, *P < 0.05; n.s.
nonsignificant; data are pooled from two independent experiments.

3. R. Eming et al., J. Invest. Dermatol. 128, 28502858

4. L. Lunardon et al., Arch. Dermatol. 148, 10311036
5. C. M. Hammers et al., J. Invest. Dermatol. 135, 742749
6. D. L. Porter, B. L. Levine, M. Kalos, A. Bagg, C. H. June, N. Engl.
J. Med. 365, 725733 (2011).
7. S. L. Maude et al., N. Engl. J. Med. 371, 15071517 (2014).
8. M. L. Davila et al., Sci. Transl. Med. 6, 224ra25 (2014).
9. D. W. Lee et al., Lancet 385, 517528 (2015).
10. J. N. Kochenderfer et al., J. Clin. Oncol. 33, 540549
11. R. J. Brentjens et al., Blood 118, 48174828 (2011).
12. C. R. Cruz et al., Blood 122, 29652973 (2013).
13. B. Savoldo et al., J. Clin. Invest. 121, 18221826
14. C. J. Turtle et al., J. Clin. Invest. 126, 21232138
15. A. Al-Amoudi, D. C. Dez, M. J. Betts, A. S. Frangakis, Nature
450, 832837 (2007).
16. B. Ohyama et al., J. Invest. Dermatol. 132, 11581168

17. K. Choudhuri, D. Wiseman, M. H. Brown, K. Gould,

P. A. van der Merwe, Nature 436, 578582 (2005).
18. S. Hennecke, P. Cosson, J. Biol. Chem. 268, 2660726612
19. K. Tsunoda et al., J. Immunol. 170, 21702178 (2003).
20. G. Di Zenzo et al., J. Clin. Invest. 122, 37813790
21. M. J. Cho et al., Nat. Commun. 5, 4167 (2014).
22. C. R. F. Monks, B. A. Freiberg, H. Kupfer, N. Sciaky, A. Kupfer,
Nature 395, 8286 (1998).
23. A. Grakoui et al., Science 285, 221227 (1999).
24. J. R. James, R. D. Vale, Nature 487, 6469
25. R. Varma, G. Campi, T. Yokosuka, T. Saito, M. L. Dustin,
Immunity 25, 117127 (2006).
26. S. J. Davis, P. A. van der Merwe, Nat. Immunol. 7, 803809
27. I. Melero et al., Nat. Med. 3, 682685 (1997).
28. Z. Ye et al., Nat. Med. 8, 343348 (2002).
29. S. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 101, 49904995
30. Z. Zhao et al., Cancer Cell 28, 415428 (2015).
31. A. H. Long et al., Nat. Med. 21, 581590 (2015).

8 JULY 2016 VOL 353 ISSUE 6295


Downloaded from on July 7, 2016










Px44 CAR


C t




32. D. A. Culton et al., J. Invest. Dermatol. 135, 15901597

33. P. Katsamba et al., Proc. Natl. Acad. Sci. U.S.A. 106,
1159411599 (2009).
34. M. Hudecek et al., Clin. Cancer Res. 19, 31533164
35. K. Watanabe et al., Blood 124, 4799 (2014).
36. S. Srivastava, S. R. Riddell, Trends Immunol. 36, 494502
37. H. Wardemann et al., Science 301, 13741377

HL087064); and the Italian Ministry of Health (G.D.Z., Ricerca

Finalizzata RF10-2309790). The content is solely the responsibility of
the authors and does not necessarily represent the official views of
the National Institutes of Health.

Materials and Methods
Figs. S1 to S11
Table S1
Movies S1 and S2
References (3855)
17 March 2016; accepted 9 June 2016
Published online 30 June 2016

Downloaded from on July 7, 2016

We thank D. Margolis, S. Prouty, T. Dentchev, C.-Y. Tsai, and

S. Nunez-Cruz for technical assistance and consultation on the
studies and J. R. Stanley for helpful discussions. The data reported
in this manuscript are tabulated in the main paper and in the

supplementary materials. Constructs and cell lines are available

from the corresponding authors under a Material Transfer
Agreement with the University of Pennsylvania. C.T.E., V.G.B.,
M.C.M, and A.S.P. have filed patent PCT/US15/28872, 2015, which
relates to compositions and methods of chimeric autoantibody
receptor T cells. C.T.E. and A.S.P. have filed Provisional Patent
Application 62/222,132, 2015, which relates to the VRC01 chimeric
antigen receptor. Research reported in this publication was
supported in part by the Penn Institute for Immunology (A.S.P. and
M.C.M.); the Dermatology Foundation Charles and Daneen Stiefel
Scholar Award (A.S.P.); the National Institute of Arthritis and
Musculoskeletal and Skin Diseases of NIH (A.S.P., R01-AR057001
and R01-AR068288; M.J.C., T32-AR007465 and F31-AR066456; G.C.,
R01-AR055309; and Skin Diseases Research Core grant P30AR057217); Deutsche Forschungsgemeinschaft (C.T.E., EL711/1-1);
the National Cancer Institute of NIH (V.G.B., T32-CA009140); the
National Heart, Lung, and Blood Institute of NIH (V.G.B., K12-


8 JULY 2016 VOL 353 ISSUE 6295 SCIENCE

Image: jim /AdobeStock

Be Among the First to Publish

in Science Robotics


Science Robotics is a unique journal created to help advance the research

and development of robotics for all environments. Science Robotics will
provide a much-needed central forum to share the latest technological
discoveries and to discuss the felds critical issues.

Join in the excitement for the Fall 2016 debut!

Dont miss the debut of

Image: Eraxion / iStockPhoto

Science Immunology.


Science is expanding its reach into immunologynow

ofering the newest online-only, weekly journal in the
Science family of publications. Science Immunology will
provide original, peer-reviewed research articles that
report critical advances in all areas of immunological
research, including studies that provide insight into the
human immune response in health and disease.

Be a part of the Science Immunology debut

issue publishing Summer 2016!
Submit your manuscript today at

AAAS seeks to recognize an individual or

a team working together in the scientic,
engineering, or foreign aairs communities
who are making an outstanding
contribution to furthering science diplomacy.





Kong-Yan Wu et al. (Zhen-Ge Luo), Semaphorin 3A activates the guanosine

triphosphatase Rab5 to promote growth cone collapse and organize callosal axon
projections, Sci. Signal. 7, ra81 (2014). Photo Credit: Rat Brain Slice. Photo Credit:
Kong-Yan Wu and Zhen-Ge Luo, Chinese Academy of Sciences.

Nominate a science diplomat before Sept 1st:

Find out more about the scope of the journal and submit
your research today.

A career plan customized
for you, by you.
For your career in science, theres only one

Features in myIDP include:

Exercises to help you examine your skills, interests, and values
A list of 20 scientic career paths with a prediction of which ones best t your skills and interests
A tool for setting strategic goals for the coming year, with optional reminders to keep you on track
Articles and resources to guide you through the process
Recommended by
leading professional
societies and the NIH

Options to save materials online and print them for further review and discussion
Ability to select which portion of your IDP you wish to share with advisors, mentors, or others
A certicate of completion for users that nish myIDP.
Visit the website and start planning today!
In partnership with:


Produced by the Science/AAAS Custom Publishing Office


Electronic Pipettes
Gilson has redesigned and expanded the
range of its easy-to-use, PIPETMAN M
line of single and multichannel electronic
pipettes. The pipettes now include guaranteed performance in repetitive pipetting mode and a programmable custom
mode, enhancing results reliability. For
sensitive biological assays like qPCR,
repeatable results depend on accurate,
precise pipetting, but until now most
pipettes have promised performance
in only one standard pipetting mode
for both standard and repetitive tasks.
Gilson now guarantees reliable perfor    
Five pipetting mode options enable
researchers to perform a large number of
applications: pipet (standard), repetitive,
mix, reverse, and custom (personalized),
with a fast access menu to all the modes
plus intuitive navigation. The new personalized pipetting mode lets researchers create their own multistep pipetting
protocols, which can be executed automatically using the simple PIPETMAN M
software interface.
For info: 800-445-7661


Liquid Nitrogen Generator

Gel Scanning
The Bio-1000F
is an innovative, userfriendly, and
device that
integrates image capture, gel preview,
and gel extraction essential for routine
nucleic acid gel electrophoresis. Using a
combination of high-sensitivity, 300-dpi
sensors and blue LED illuminators, the
Bio-1000F is compatible with all ethidium
provides publication-quality images. With
a sensitivity of up to 0.04 ng per band, the
       sults over the more traditional CCD-based
gel documentation systems. Users simply
place their gel onto the scanning area,
close the lid, and wait for the image to be
captured. A convenient orange screen
shields the user from excessive exposure to the blue light. Users can visualize
band patterns and conduct gel extraction
directly on the Bio-1000F. Less than 30
cm wide, the Bio-1000F has a compact
small space in a crowded laboratory.
For info: +44-(0)-1223-515440

The new Metrohm 917 Karl Fischer

Coulometer is a compact, stand-alone

features that make it ideal for fast,
precise determination of low-level water
content. A vivid touchscreen display,
simple method recall, and an integrated
pump combined with reagent and sample handling make the
917 compact coulometer the fastest, most convenient solution
for routine moisture analysis. The 917 is intelligent, automatically
recognizing when a sample has been injected and starting the



with automatic sample-weight transfer from a laboratory balance,
and laboratory information management system compatibility.
Quick access to frequently used methods allows titration to begin
with a single touch on the high-resolution color touchscreen.
Furthermore, the 917 is compact, requiring only 8 in. of linear
bench space. The coulometer can be connected to an oven for
thermal sample preparation, enabling it to support the broadest
range of sample types.
For info: +41-(0)-71-353-85-85

The Asynt CryoSyn is a unique, ondemand liquid nitrogen (LN2) and nitrogen gas generator (available in 10-L,
20-L, 40-L, 60-L, and 120-L capacity
models) for industrial, medical, food and
agricultural, biological, and general laboratory applications. CryoSyn eliminates
the need for storage and transportation
of LN2, thereby making its use more
convenient, reducing running costs,
and eliminating traditional evaporation/
contamination problems. The CryoSyn
control system gives operators all the
information necessary to ensure an ef

   2 and
nitrogen gas. Designed to ensure high
operational safety, the CryoSyn operates
automatically and provides information

single-phase electrical supply, the 10-L,
20-L, and 40-L CryoSyn units are fully

 way, enabling easy transport between

of investment in less than two years.
For info: +44-(0)-1638-781709


The RFM-T Series of high-precision Peltier temperature-controlled benchtop

refractometers features a wide measuring range of up to 1.70 RI (refractive
index), precision to six decimal places
RI (three decimal places Brix), and an
all-new, 7-in. touchscreen user interface.
The RFM-Ts are ideally suited for quality control applications within
the food and beverage, chemical, petrochemical, pharmaceutical,
!  ! 
   !      ! 

! plication. RFM900-T models cover the widest of measurement
ranges and incorporate a Kalrez prism gasket to protect the instruments from the harsh chemicals at the upper end of their refrac


for easy cleaning between viscous samples. The touchscreen and
graphic user interface help the operator quickly navigate through
the menus. On-screen graphical prompts support method loading,
calibration, and routine maintenance.
For info: 978-778-1010

Electronically submit your new product description or product literature information! Go to for more information.
Newly offered instrumentation, apparatus, and laboratory materials of interest to researchers in all disciplines in academic, industrial, and governmental organizations
are featured in this space. Emphasis is given to purpose, chief characteristics, and availability of products and materials. Endorsement by Science or AAAS of any
products or materials mentioned is not implied. Additional information may be obtained from the manufacturer or supplier.


8 JULY 2016 VOL 353 ISSUE 6295


want new

Learn about the
latest breakthroughs, new
technologies, and
research in a variety
of felds. Our
expert speakers
explain their quality
research to you and
answer questions
submitted by live




Brought to you by the Science/AAAS

Custom Publishing Ofce








e c t i o n yo u c


R&D Systems is Your New Trusted

Source for Luminex Instrumentation!

Direct access to Luminex staff for world-class

on-site and 24x7x365 remote support

Unlimited emergency equipment repair

Discounts and special acquisition programs
tailored to any labs budget

Learn more about Luminex instruments |

Global TEL +1 612 379 2956

North America TEL 800 343 7475 Europe | Middle East | Africa TEL +44 (0)1235 529449
China TEL +86 (21) 52380373


Theres only one

Science Careers

your career
with expert
advice from

For full advertising details, go to and click
For Employers, or call one of
our representatives.
Tracy Holmes
Worldwide Associate Director
Science Careers
Phone: +44 (0) 1223 32()2)

Fax: +1 (202) 28+ (742
Tina Burks
Phone: +1 (202) 32( ()77
Nancy Toema
Phone: +1 (202) 32( ()78
Online Job Posting Questions
Phone: +1 (202) 312 (37)


Fax: +44 (0) 1223 32()32
Sarah Lelarge
Phone: +44 (0) 1223 32()27
Kelly Grace
Phone: +44 (0) 1223 32()28
Online Job Posting Questions
Phone: +44 (0) 1223 32()28

Katsuyoshi Fukamizu (Tokyo)
Phone: +81 3 321+ )777
Hiroyuki Mashiki (Kyoto)
Phone: +81 7) 823 110+


Ruolei Wu
Phone: +8( 18( 0082 +34)
Danny Zhao
Phone: +8( 131 4114 0012
All ads submitted for publication must comply with
applicable U.S. and non-U.S. laws. Science reserves
the right to refuse any advertisement at its sole
discretion for any reason, including without limitation
for offensive language or inappropriate content,
and all advertising is subject to publisher approval.
Science encourages our readers to alert us to any ads
that they feel may be discriminatory or offensive.

Download Free Career Advice Booklets!
Featured Topics:
Industry or Academia
Job Searching
Non-Bench Careers
And More

Join us !
Science & Diplomacy
provides an open access forum
for rigorous thought, analysis, and
insight to serve stakeholders who
develop, implement, or teach all
aspects of science and diplomacy.
Learn more about the latest ideas in
science diplomacy and receive regular
updates by following @SciDip on
Twitter, liking the quarterlys page
on Facebook (
sciencediplomacy), and registering
for free at www.sciencediplomacy.

We are looking for an experienced, entrepreneurial and passionate

scientifc leader to join our Neuropsychiatry department as the:
This role is truly essential to lead and coordinate, with a clear
industrial vision, our Research activities from the concretization
of promising new targets through the drug discovery phase to early
clinical development.
Reporting directly to the Head of Neuropsychiatry R&D, you will
be responsible for orchestrating our Proteinopathy programs for the
treatment of neurodegenerative disorders.
Responsible for the strategic progression of the Research Projects
Portfolio, you will offer your scienti&c insight both into in-house
projects and into potential licensing-in opportunities.
Collective Intelligence is a critical pillar of future progress and,
as Program Director, you will help build our future success by
facilitating and structuring exchanges between the highly-engaged
Project Directors, translational and clinical teams, experimental
platforms and external collaborators.
Major requirements:
Ph.D. in neuroscience
10 years research in an academic environment
Previous experience in a pharmaceutical &rm and/or in drug
Established reputation in the &eld of neurodegenerative disorders
Specialist in the domain of Proteinopathies
Collaboration-based mindset and excellent interpersonal skills
Perfectly fuent in English - French would be a plus
Joining Servier means working in a stimulating environment and
contributing to an innovative, research-based and people-oriented
Servier is the leading French independent pharmaceutical company,
located in 144 countries with 21 000 employees and a turnover in
2015 of about 4 billion euros.
25% of Serviers annual turnover is reinvested in Research &
Development, refecting the companys dedication to its mission of
innovation and discovery for improved treatment of unmet medical
Organized as a foundation, we guarantee our independence toward
stock exchange market. Our founder Doctor Jacques Servier used
to say: a company is a group of human beings, so it has no price.
Servier is well known for staff care, with particular attention to career
evolution within the Group, and to an agreeable and productive
working environment.
Science & Diplomacy is published by the Center for Science
Diplomacy of the American Associaton for the Advancement of
Science (AAAS), the worlds largest general scientfc society.

Together with international experts, for over 30 years we have been

investing in programs to improve the control of neuropsychiatric and
neurodegenerative diseases like Alzheimers, Parkinsons, multiple
sclerosis and depression.
Our Research Center in Neuropsychiatry is located near Paris, in
Croissy-sur-Seine, France.
To apply or to acquire more details concerning the position, please
contact Geoffroy BELLET via
- ref.965


Institut de Recherches Servier

ways that Science Careers

can help advance your career

Register for a free online account on


Search thousands of job postings and nd

your perfect job.


Sign up to receive e-mail alerts about job

postings that match your criteria.


Upload your resume into our database and

connect with employers.


Watch one of our many webinars on

dierent career topics such as job
searching, networking, and more.


Download our career booklets, including

Career Basics, Careers Beyond the Bench,
and Developing Your Skills.


Complete an interactive, personalized

career plan at my IDP.


Visit our Career Forum and get advice from

career experts and your peers.


Research graduate program information

and nd a program right for you.


Read relevant career advice articles from

our library of thousands.

today all resources are free


Review of applications will begin

immediately and will continue until the
position is flled.
For complete qualifcation/application
information, see
employment/job-list-faculty.php or
The University of Tennessee is an EEO/
AA/Title VI/Title IX/Section 504/ADA/
ADEA/V institution in the provision
of its education and employment
programs and services.

The Vaccine and Infectious Disease Division (VIDD) of the Fred Hutchinson Cancer Research Center
seeks exceptional applicants for a full-time senior faculty leadership position in immunology at the Full
Member rank (comparable to Professor). The primary responsibility of this position will be to develop and
lead a comprehensive, cross-disciplinary integrated research center (IRC) involving multiple investigators
that will focus on pathogen-induced cancers. The candidate will be expected to conduct laboratory-based
translational immunology research as part of this IRC and within VIDD, and an emphasis on mechanisms
of memory/effector cell induction and immune dysfunction in the context of cancer or cancer-associated
infections is highly desirable.
Candidates for this position must have a well established, robust, funded program that is nationally and
internationally recognized for excellence in immunology, immunotherapeutic design, or viral oncogenesis.
Applicants must have an MD (or foreign equivalent) or PhD (or foreign equivalent). Selection criteria include
excellence in scholarship, creativity in research, and demonstrated leadership in the profession.
VIDD scientists integrate clinical care, computational methods, and basic science research in immunology,
virology, and vaccine design to reduce the global burden of infectious disease.
The Fred Hutchinson offers a vibrant intellectual environment within a beautiful, lakeside campus in
Seattles South Lake Union biotech hub. VIDD occupies a new building that is connected by walking trails
to Seattle Cancer Care Alliance and the other four Divisions of the Fred Hutch and by trolley to major
research partners such as the University of Washington School of Medicine, Seattle Childrens Research
Institute, Center for Infectious Disease Research (formerly Seattle Biomedical Research Institute), and the
Infectious Disease Research Institute.
Interested candidates should submit a CV, a concise research plan statement, and the names and contact
information for three (3) references to: Specifc inquiries can be directed to Julie
McElrath at 206-667-1858. Applications should be received by August 15, 2016 to assure consideration
and will be evaluated as received.
The Fred Hutchinson Cancer Research Center is an Affrmative Action, Equal Opportunity Employer.
All qualifed applicants will receive consideration for employment without regard to, among other
things, race, religion, color, national origin, sex, age, status as protected veterans, or status as qualifed
individuals with disabilities. We strongly encourage applications from women, minorities, individuals
with disabilities and covered veterans.

Produced by the Science/AAAS Custom Publishing Oce.

Cancer Feature:
Postdoc Feature:

Issue date: August 26
Book ad by August 9 to
guarantee space
For recruitment in science, theres only one Science.
Why choose this Postdoc Feature for your advertisement?
Relevant ads lead o the career section with a special Postdoc banner.

Expand your exposure by posting your print ad online:

Link on the job board homepage directly to postdoc positions
Dedicated landing page for postdoc positions.
Produced by the Science/AAAS Custom Publishing Oce.


To book your ad:

The Americas: 202 O26 6P82
Europe/RoW: +SS 0 B22O O26P00
Japan: +8B O O2B9 PTTT
China/Korea/Singapore/Taiwan: +86 B86 0082 9OSP

Ads accepted until August B9

if space allows


The College of Dentistry at the University

of Tennessee Health Science CenterMemphis invites applications for the
position Assistant Professor (PIN
#22932) that is a 12-month tenure
eligible appointment in the Department of
Bioscience Research. The ideal candidate
for the position will be an investigator
with an innovative and actively funded
research program that is related to omics
research (e.g., proteomics, metabolomics,
genomics, transcriptomics, etc).

Senior Faculty Leadership Position in Immunology at

Fred Hutchinson Cancer Research Center

By Carlos A. Aguilar-Trigueros

The questions that opened doors

I got her name from the German

Embassyone of many embassies
some friends and I visited to inquire
about scholarship opportunities for
conducting research abroad. She
had just nished her Ph.D. in Europe and moved to El Salvador with
her husband, a physics professor at
our university. I was excited to talk
to someone who had done research
abroad, so I visited her at her ofce.
After a few discussions, she came
up with the idea of starting a journal club for students looking for
topics for their bachelors theses.
The plan was to meet every Friday afternoon to discuss papers on
plant-fungal interactionsher eld
of study, which I soon decided to
pursue because her enthusiasm was
so contagious. The journal club was
a transformational experience. I
had been trained to understand that what I read in a paper
should be taken as truth, but now, for the rst time during
my academic career, my classmates and I were deconstructing papers and identifying their strengths and aws. This
was denitely not the approach to science that my professors had taught.
Reading papers that my new mentor had written during
her graduate studies and those of people she knew also gave
the science a human side. Before, scientic articles were
just sheets of paper bearing cryptic messages, written in
a foreign language by people I didnt know who worked at
distant universities. Now, I could imagine the people behind themand even imagine being one of those people
myself. Her stories about her experiences as a graduate student were also enlightening. In El Salvador, there are no
doctoral programs in the natural sciences or mathematics,

so I was fascinated to learn that one

could dedicate a lifetime to generating questions and collecting data to
answer them.
The journal club and my other
discussions with this scientist
made the possibility of becoming
a researcher feel more tangible
and helped give me the courage
to continue with biology. Instead
of abandoning my interest in the
natural world and changing my
major, I sent emails to professors
and programs overseas, seeking a
chance to pursue graduate studies.
Those emails contained the good
enough questions that helped me
secure an amazing internship at
the Smithsonian Tropical Research
Institute in Panama, where I had
more time and training to rene
my questions. From there, I moved
on to my challenging but rewarding doctoral studies at the
Free University of Berlin.
Now, as a postdoctoral fellow, I am working to solidify my
science career. It is too soon to know where it will take me,
but I hope to somehow help improve the research and training environment in El Salvador. In the meantime, Id like to
ofer some encouragement to potential scientists in countries
like mine, and pass on thanks to the mentors who help our
research dreams become reality. Everybody can do science,
but in countries like ours it requires a little extra creativity,
perseverance, and patience to nd the proper environment
to develop your skills. Keep questioning.

Downloaded from on July 8, 2016

couple years ago, I asked my Ph.D. adviser why he decided to give me a position in his lab. His
answer was simple. Because you were so focused on nding good questions, he told me. I found
this answer amusing, because my early scientic training hadnt fostered this skill at all. In my
undergraduate years at the University of El Salvador, the biological world had been presented as a
series of known facts, and students were not trained to question them. There is little research tradition in El Salvador, and my professors didnt provide examples of what scientic inquiry is really
about. I had entered university wanting to become a scientist, but my commitment was fading; my studies began to seem pointless. Then, late in my undergraduate studies, my path to a research career opened
when I met a scientist who taught me about asking questions and showed me the human side of science.


Carlos A. Aguilar-Trigueros is a postdoctoral fellow at the

Free University of Berlin. Send your science career story to SCIENCE

8 JULY 2016 VOL 353 ISSUE 6295

Published by AAAS


I was fascinated ... that one

could dedicate a lifetime to
generating questions.